GB2534882A

GB2534882A - Apparatus, method, and program for predicting vehicle collision rates

Info

Publication number: GB2534882A
Application number: GB1501737.9A
Authority: GB
Inventors: Albornoz Jose
Original assignee: Fujitsu Ltd
Current assignee: Fujitsu Ltd
Priority date: 2015-02-03
Filing date: 2015-02-03
Publication date: 2016-08-10
Anticipated expiration: 2035-02-03
Also published as: GB2534882B; GB201501737D0

Abstract

A collision prediction apparatus 10, the apparatus comprising a simulator 14; the simulator being configured to execute an iterative agent-based simulation in which, in each iteration: a subject agent representing a travelling party (e.g. pedestrian, motor vehicles, bicycle) travels from an end of a simulation environment representing a thoroughfare toward another end; one or more further agents representing vehicles interact with the subject agent, the simulator being configured to categorise the outcome of each interaction as either a collision or not a collision; the iteration being complete when the subject agent either reaches the another end, or is involved in an interaction of which the outcome is categorised as a collision; the simulator being configured to record and output a collision rate value representing a number of interactions in which the outcome is categorised as a collision per iteration. The apparatus may be configured to use real-time data provided by a monitoring module 20 representing the traffic on a particular section of road in the real-world domain. The apparatus may generate an output alert which may cause a signage located on a road to display a warning or a warning may be generated within a vehicle.

Description

Apparatus, iMethod, and Program ctsng Vehicle Collision Rates present lies in the field of intelligent transport teehttlogy. Specifically, the invent to a procedure for s chicle behaviour o predict road accident Road traffic accidents are a sign ant cause of sc and fatalities. In addition_ road traiC s are a major of congestion, disruption, and a significant use of emergency service and health service resources. In particular, sections of road which are subject to mixed use by different categories of vehicle can pose a significant risk to slower moving vehicles, for example, bicycles. Every year some 19,000 cyclists are killed or injured in reported road accidents in the UK alone, and according to the US Department of Transportation, 677 cyclists were killed in motor vehicle accidents in 2011, while 48,000 were injured. Cyclist casualties have risen in recent years as the amount of cycling has increased, with the number of journeys made by bicycle in Greater London having doubled between 2000 and 2012 to over 540,000 per day; in November 2013 alone six cyclists were killed on London streets within a two-week period, bringing the number of cyclists killed in London in the year to 14.

Amongst the road events associated to cycling accidents overtaking poses the not only because it can put the overtaking vehicle in the path of oncoming traffic but because it increases the chances of a cyclist being hit by a vehicle, often at high speeds. Moreover, researchers from the University of Bath and Brunel University published experimental results suggesting that there is very little a cyclist can do to prevent vehicles from overtaking in a dangerous manner: measured average overtaking distances between cyclists and passing vehicles showed little variation in spite of high-visibility outfits, and in the most dangerous overtakes passing distances were as small as 50 cm.

It is desirable to identify locations at Inc') accidents e likely to occur, and en mental and traffic conditions under hich there is particular danger.

Embodiments include a collision prediction apparatus, the apparatus comprising a simulator; the simulator being configured to execute an iterative agent-based simulation in which, in each iteration: a subject agent representing a travelling party travels from an end of a simulation environment representing a thoroughfare toward another end; one or more further agents representing vehicles interact with the subject agent, the simulator being configured to categorise the outcome of each interaction as either a collision or not a bjeet agent either reaches the another end, or the simulator being con er which the outcome is categorises hei enaction esf which t and output a co a collision per iterati The embodiment provides an in-silico simulation platform that generates synthetic road traffic statistics agent-based model of tr elling parties interacting on a thoroughfare. The category of represented by the subject agent is configurable depending on the implementation nay be, for example, a bicycle, whereas the further agents may represent motor vehicles, Alternatively, veiling agent may be a pedestrian, and the vehicles represented by the further agents bicycles and/or motor vehicles. An agent in the simulation representing a travelling pang means that properties attributed to the agent in the simulation (for example, by a controller in an initiation phase) are selected from a range of values that the travelling party may possess in the real world domain. Such ranges may be derived by monitoring thoroughfare usage and collecting statistics, or may be input as arguments to the simulator by a controller or an administrator. The properties attributed to agents include one or more of the following: dimensions, shape, speed, acceleration, deceleration, and manoeuvrability, and visibility, among others. Visibility may also be partially dependent upon factors represented by the simulation environment. Factors attributed to agents affecting visibility may include, for cyclist agents, presence of lights, brightness of lights, and whether or not high visibility clothing is worn. For a motor vehicle, visibility may be affected by a colour of the bodywork of the motor vehicle and/or light brightness. In other words, values of properties attributed to agents in the simulation are set at values corresponding to real-world values of the same properties of the travelling party being represented. Such real-world values may be obtained from historical databases, through real-time observation, or by other means.

25.A travelling party is taken to be an umbrella term including vehicles and pedestrians. Thus, each agent represents a travelling party. However, further agents, which specifically represent vehicles, do not represent pedestrians. A vehicle containing passengers is taken to be a single travelling party for the purposes of this document. Properties attributable to a driver or operator of a vehicle may be discussed as being properties of the vehicle, this is for ease of discussion, and does not imply that vehicles possess human characteristics.

Agents is taken to he a collective term for the subject agent and the further agents. Other agents is taken to be all agents (whether they he the subject agent or further agents) other than the particular agent being discussed. Furthermore, when decisions, determinations, or assessments are described as being made or taken by an agent, it is assumed tha the simulator performing the iteration has made the determination/decision/assessment using rules attributed to the agent.

The simulation taken to mean the spatial runcnt hr which the agents move, representing a thoroug e, properties ncluding the dimensions and features of the layout (including, example, number of lanes, direc vet lane, speed limits, road markings and their implications on vehicle behaviour, raised features segregating areas of the road from one another, junctions, signage, lighting) of a thoroughfare, and possibly its surrounding environment, are attributed to the simulation environment at values corresponding to either a particular section of thoroughfare in the real-world domain, or to database values representing typical or proposed thoroughfare layouts. An additional factor that may be represented by the simulation environment is visibility. Visibility (or vehicle visibility) may be affected by factors including time of day, and weather conditions, and may be represented by, for example, a distance. Thoroughfare is taken as an umbrella term covering all surthces along which parties travel and interact. In particular implementations, the term thoroughfare may be limited to roads and/or paths. In particular, the thoroughfare being represented by the simulation environment may be a road.

The movement of agents in the simulation environment rule-based and responsive to the dynamic geospatial situation in the simulation environment. That is to say, the relative position of elements of the simulation environment (for example, the edges of the thoroughfare) and position and direction and speed of travel of other agents compared to the agent in question determine how the agent in question moves within the simulation environment. Each agent may be considered to be autonomous in terms of there being a set of rules for each agent in each simulation, and the application of those rules to the geospatial situation in the simulation environment in order to determine how the agent moves effectively forms a single processing thread. However, the agents interact with one another, at least insofar as the movements determined for one agent being at least partially dependent on the relative position and absolute or relative velocity of other agents.

At any' point in time during the te involved in the iteration at that point in time occupies a space in the simulation environment. The extent of the space occupied by each agent depends on the properties attributed to the agent and hence to the travelling party represented by the agent. The outcome of an interaction may be categorised as a collision if, at any point in time during an iteration, any point in the space occupied by one agent in the simulation environment overlaps or becomes coincident with any point in the space occupied by another agent. Alternatively or additionally, a distance-speed threshold could be defined such that the outcome of an interaction is categorised as a collision if the distance between each agent example, motor vehicle agent and a subject agent such as a cyclist agent) is less old while the motor vehicle agent's speed exceeds a speed threshold (a speed-fiber alternative, each time the distance-speed threshold is satisfied a probability mild be used to determine whether or not a collision is deemed to have occurred. The term "collision" may therefore include events that imply a threshold associated with a particular likelihood of an dent rring has been exceeded, and not only those events which imply physical contact. An two agents may be considered to begin when either agent begins to include the position and/or velocity of the other agent in determining how to move, and ends when neither agent includes the position and/or velocity of the other agent in determining how to move, or when the iteration ends.

Furthermore, regardless of whether or not either of a pair of agents includes the position and/or velocity of the other agent in determining how to move, two agents whose movements are such that they will collide if they continue on their present paths are also considered to be interacting.

The simulator executes an iterative simulation. Thus, a single simulation comprises a plurality of iterations.

The number of iterations per simulation will depend on the implementation details of the embodiment and also on the requirements for each simulation. It may be that a controller or some other component of the apparatus, or third party, requesting/instructing the simulation specifies the number of iterations in the simulation. In each iteration, the subject agent either travels from one end of the simulation environment to another end of the simulation environment safely (that is, without being involved in an interaction with another agent which results in a collision), or is involved in a collision (that is, is involved in an interaction with another agent, the outcome of the interaction being categorised as a collision). After a number of iterations, a proportion of those iterations which ended with the subject agent being involved in a collision is output, either to the component or third party requesting/instructing the simulation, or otherwise. Depending on the particular implementation, it may he that only interactions involving collisions in which the subject agent collides with another agent are categorised as collisions, or it may be that any interaction involving the subject agent which results in two agents colliding is categorised as a collision.

In the real-world domain, an interaction between ehicles which pose a particular risk of accident is a faster moving vehicle overtaking a slower moving vehicle. Therefore, such interactions are of particular interest in a simulation. Since, in the real-world domain, a faster-moving vehicle approaching a slower-moving vehicle from behind will not always overtake, but may slow down and stay behind, the simulator may be configured to set up at least one potential overtaking situation per iteration, and the agent-based model attributed to the agents involved, respectively, will determine whether an overtake occurs or whether the further agent slows down and stays behind the subject agent. In particular, in each iteration, the one or more further agents interacting with the subjectagent includes one lore further agents being positions away from the end of the simulation environment toward which the subjectagent is moving, the further agent moving toward the same end of the simulation environment subject iwhich, if maintained, would result in the further agent reaching the en agent _. environment before the subject agent.

In other words, in each iteration, at least one furthe nt approaches the s using the rules attributed to the at least one further agent, determi behind, and 'crake the subject agent, hence, the further agent is referred to as a potential overtaking agent. Of course, the determination may also include at what distance behind the subject agent to begin an overtaking manoeuvre (moving outside of the subject agent in order to pass) if one is necessary (i.e. it may be possible to overtake without a specific manoeuvre, that is, without changing velocity), and by what lateral separation to pass the subject agent.

Of particular interest is motor vehicles overtaking, or approaching from behind at a faster speed and hence having the potential to overtake, bicycles, since that is a cause of a significant number of accidents involving bicycles. Hence, embodiments include those in which the subject agent represents a bicycle and, in each iteration, the one of the one or more further agents involved in the interaction defined above is a motor vehicle.

Each agent is attributed a set of rules governing the behaviour of that agent, wherein behaviour is effectively movement within the simulation environment, movement including changing velocity both longitudinally (i.e. in the direction of travel: speeding up or slowing down) and laterally (i.e. moving from lane to lane, moving out to pass another agent, and moving back in after passing another agent).

In one example, the movements of each of the agents within the simulation environment are determined by a set of rules attributed to the agent by the simulator during an initiation phase of each iteration, the set of rules governing geometrical assessments used by the agent during interactions with other agents, and consequential movements based on the geometrical assessments.

The initiation phase of each iteration may be performed by the simulator, or may be performed by a controller component of the collision prediction apparatus. Alternatively, the initiation phase may he undertaken by an administrator or user, or may be at least partially configurable by inputs provided by an r^ n n CI 1,r1C1 at initiation phase may med per simulation, or there may be one initiation phase per A geometrical -ment may he a determination by the agent of whether spaces between elements of the simulation environment are sufficient to render a pa cement feasible ssessinent may be made in the light of a particular geometrical scenario, and the agent-based iount the probability of error on the part of the agent (or the operator of a vehicle in the real-world domain), so that the geometrical assessment made by the agent may not necessarily correspond to the true geometrical scenario in the simulation environment.

Particular examples of geometrical assessment include: longitudinal and/or lateral distance between the potential overtaking agent and the subject agent; longitudinal and/or lateral distance between the potential overtaking agent and any oncoming agent.

Such geometrical assessments may be undertaken by the potential overtaking agent, by the subject agent, and/or by one or more other agents. When a geometrical assessment is described as being undertaken by a particular agent, it is taken to mean that it is assessed using rules attributed to the particular agent, and the result used as an input to a rule determining whether and how the particular agent is to change velocity.

The agent-based rules for determining how the agents interact one another may be implemented such that each of the geometrical assessments performed by an agent gives rise to a result which is used by the agent as an input to one or more rules governing which consequential movement to make, and wherein [he result of the geometrical assessment is calculated by randomly selecting one value from a range of values ncluding the true/actual value of the respective distance in the simulation, A consequential movement in this sense may be a change in either or both of lateral and longitudinal velocity, or a change in lateral position. The geometrical assessment may not be in terms of a distance or separation, but may be in terms of whether the geometrical situation in the simulation environment at the time of the assessment makes a particular movement by the agent feasible (that is, able to occur without requiring another agent to change velocity in order to avoid a collision).

By observing real world behaviour of vehicles moving on roads, probability distributions can be established which represent the behaviour of vehicles, and such probability distributions can be utilised both when calculating the result of a geometrical assessment, and when applying rules which use the result of the r..-1 ennnne net a how the agent moves. That is to say, iorld behaviour of vehicles, probability distributions representing the response of a vehicle) a set of geometrical tin terms of likelihood to respond in each of a predetermined ntnuber can be probability distributions be used as a basis for the rules governing agent hav i our, st ibutions may he made specific to particular attributes of the vehicle, for ample, vehicle type (HGV/car/bicyclehnotorbike), driver/operator gender (and/or age), the observed real world behaviour may be historical and stored in a database included in or accessible to the simulator, or may include real time data obtained from measurements made by roadside monitoring equipment (either at a particular location represented by the simulation environment or at a number of locations having characteristics in common with the simulation environment).

Alternatively, the result of the geometrical assessment s calculated by adding a random error to thevalue representing the respective distance in the simulation.

The random error may be selected according to a probability distribution, so that certain values are more likely to be selected than others. Furthermore, the random error may he selected from a range of values (according to a probability distribution distributed around the actual value of the respective distance in the simulation) wherein the extent of the range of values is proportional to the speed of the motor vehicle agent. Thereby the difficulty of judging distances at high speeds is simulated.

The one or more rules governing which consequential movement include whethe when, at what speeds, and by what lateral and longitudinal margin, for the potential overtaking agent to overtake the subject agent.

Optionally, at the initiation of each teration or each simulation, or indeed whenever a new agent is introduced to a simulation at an end of the simulation environment, the subject agent and/or one of the one or more further agents are allocated a value representing attitude during an initiation phase of each iteration, and the outcome of applying the one or more rules governing which consequential movement to make is at least partially dependent upon the value representing attitude.

Alternatively or additionally, the subject agent and/or each of the one or more further agents arc allocated a value representing attitude during an initiation phase of each iteration, and the set of rules attributed to the agent by the simulator during the initiation phase are configured according to the value representing attitude.

8 be represented by a dynamic mathematical model It ively, or additionally, driver,a the plurality of factors that takes into account a plurality of factors influent. n conditions.

Tla or whether an including one or any combination of age, Eype of vehicle, time of day, an single value representing attitude is allocated to the driver in a probabilistic represented by the dynamic model outlined above, can vary between implementat is that the dynamic model is utilised to calculate the value representing attitude. A further alter Another factor whose impact is harder to quantify is the psychology of the polite, G. drivers will tend to wait for oncoming traffic to pass by before overtaking a cyclist, while rude, selfish drivers will overtake a cyclist regardless of oncoming vehicles. It would therefore be desirable to have a way to incorporate the effect of these behaviours in cycling accident statistics in order to measure the possible impact of driver awareness campaigns. Hence, the attitude value may be randomly allocated to each agent, or a proportion of drivers having each attitude value may be configurable at the instruction of a simulation. In a simple example, there may be only two values: selfish or unselfish. The rules may be configured according to the value representing attitude by making certain movements more likely in response to a geometrical assessment, or by precluding the driver from certain manoeuvres altogether in the ease of having a particular attitude value. The attitude values may be used to configure rules by setting the likelihood of the agent to execute a particular consequential movement in response to a given set of geometrical circumstances in dependence upon the value representing attitude.

A controller, or control module, may L dcd in the apparatus as an entity responsible for providing arguments (that is, values of input parameters) to the simulator, with the simulator being responsible for the procedure of performing the simulation according to the arguments provided. For example, the controller may receive instructions and input parameters from a third party requiring a prediction of collision rates.

The controller may include a GUI for obtaining input parameters from a user, As a further alternative, input parameters may have default values whieh the simulator uses if no value is specified by the controller.

An exemplary collision prediction apparatus includes: a controller configured to specify a set of input parameters and to instruct the simulator to execute a number of iterations with the specified input parameters, wherein the input parameters include one or more of the following: the dimensions and/or layout of the road represented by the simulation environment; the visibility conditions in the simulation environment; the density of agents in the simulation environment and/or the average speed of the agents; a proportion of further agents belonging to each of a predetermined set of categories of agent, including bicycle, car, IIGV, and/or motorcycle; a category of agent to which the subject agent belongs from among a predetermined set of categoriesHOY visibility of the subject agent and/or factors determining visibility including clothing righ if the subject agent belongs to the category of bicycle, other characteristics including gender of rider, and whether or not the rider is wearing a helmet; wherein the set of rules attributed to each agent in the s configurable according to the input parameters.

The above input parameters all relate to the agent-based modelli simulator. and hence are used within each iteration. However, input parameters may also he general instructions relating to the form of the simulation, such as how many iterations to perform or how long to run for, which features of the simulation to observe and record, and how to output the results. Where visibility is a factor, it may be that, in order to determine the distance at which one agent is visible to another, the visibility attributed to the one agent is combined with the visibility attributed to the simulation environment. For example, the visibility attributed to the simulation environment may be represented by the maximum distance at which another agent is visible in the given conditions, and then each agent attributed a value of 0 to 1 representing visibility with 0 being invisible and 1 being peak visibility (i.e. the most visible colour possible with the brightest lights possible). The visibility of an agent (i.e. the distance at which the agent becomes visible to another agent) could then be determined by multiplying the visibility attributed to the simulation environment by the visibility attributed to the agent.

Any of the above input parameters may d by a value from a predetermined range of values defined for the parameter. One or more of the input parameters may he selected to represent a feature of the real world domain (for example, road width or maximum speed limit), and optionally, input parameters selected to represent a feature of the real world domain may be modified in order to investigate the impact of changing those features.

Certain of the input parameters, such as vehicle density and wherein the average speed, may be on a per-category of vehicle basis. Categories of vehicle may be simply bicycle and motor vehicle, or motor vehicle may be further divided into two or more of motorbike, HGV, van, car.

The agent-based rules being configurable according to the input parameters means, for example, that the data on which the rules are based (and in particular any probability distributions on which the rules are based) are selected to reflect the input parameters. Agents belonging to a motor vehicle category may respond differently to an agent belonging to the category of bicycle in dependence upon whether or not the rider of the bicycle is wearing a helmet. For example, in a rule determining whether or not to move out to overtake a bicycle agent, and how much lateral space to allow, the rule may be configured so that a car agent is less likely to overtake if the rider of the bicycle agent is not wearing a helmet, and may tend to allow more lateral space the overtake.

ntcrventionsucho bicycle lanes or wide enough central separations could contribute to a reduction Mg accident statistics is no well-es methodology to evaluate the cost/benefit ratio of such measures and their -without actually having to implement them.

Embodiments may be utilised as a means of selecting modifications to make to thoroughfares in the reaE-world domain. For example, with all other input parameters being fixed, candidate real-world implementation options such as signage (such as max/recommended speed) or road layout (marked bike lane/no marked bike lanehnore lanes/fewer lanes) can be represented in the simulation environment and the collision rates simulated with each implementation option or combination of implementation options represented. The implementation option or combination of implementation options giving rise to the lowest collision rate can be output as a design selection.

As a particular example, the apparatus further comprises a road layout selection module, being configured o generate the simulation environment in a plurality of versions, each version differing from one another in dimensions and/or layout, and to output the generated simulation environments to the simulator; the simulator being configured to execute a predetermined number of iterations for each version of the simulation environment, with any other input parameters being: the same for each version; the road layout selection module being further configured to receive the value output by the simulator for each version of the simulation environment, and to identify and output the version of the simulation environment having the lowest collision rate as a suggested road layout.

Cost may also be taken into account, so that each of the versions of the simulation environment is associated with a value representing implementation cost, and the road layout selection module is configured to compare the candidate road layouts using a metric based on the respective output collision rate values and the respective values representing implementation cost, and to identify and output the version of the simulation environment scoring highest by the metric.

Alternatively ditionally, the collisa omprise a configurable signage control module, wherein the simulatio environment represents a particular section of road in the real world domain which features configurable s simulator is instructed to execute simulations with all input parameters fixed other than differen d speed limits from a predetermined set of candidate limits, the configurable signag of module being configured to receive the collision rate value output by the simulator for each candi ate limit if the collision rate values output for one of the candidate limits satisfies a selection criterion, to send a control signal to the configurable sin implement the one of the candidate limits for which the collision rate value satisfies the selection criterion. An exemplary criterion is less than a fixed proportion (Le. 035 or 0.5) of the collision rale value output for the candidate limit corresponding to the limit currently displayed on the configurable signage.

The simulator performs the simulation in-silitio The interactions being simulated are geometric in their nature, and hence the simulations are of significance when displayed visually. This may be useful for a user to understand why particular input parameters have the effect that they do on the output collision rate value.

Optionally, the apparatus further comprises a graphical representation generator configured to generate and output a graphical representation of the simulation environment and the agents in the simulation environment.

Embodiments may further comprise a communications module configured to generate and output an alert signal when the collision rate value output by-the simulator satisfies an alert criterion.

Partic lions when the simulation environment represents a section of thoroughfare in the real-world domain, if the collision rate value exceeds a threshold (as an example of an alert criterion) it may be that an alert signal can be used as a means to alert an authority responsible for highway management that accidents are likely, or that the alert signal be linked to some configurable signage on the road to decrease speed limits or to display a warning to thoroughfare users.

Potential uses of the collision prediction apparatus extend to processing real-time data in order to alert users of a particular section of road that conditions are such that there is an enhanced likelihood of an accident occurring. For example, the simulation environment may represent a particular section of road, and real-time data representing the density of vehicles on the particular section of road and/or the average speed of the vehicles is monitored by a monitoring module, and the controller is configured to input the real-time data to the simulator as input parameters, the simulator being configured to execute a simulation in which the density of agents and/or average agent speeds in the simulation correspond to the real-time data, and the simulation environment represents the particular section of:it'd; and the alert sign configurable 2,C located at the particulat it to ring.

-native to being a roadsideapparatus, the collision prediction apparatus may be provided as a component of a motor vehicle, for example, the ion prediction apparatus is a component of a vehicle; the collision prediction apparatus further comprises a real-time data receiving module configured to receive real-time data representing the density of vehicles on a particular section of road and/or the average speed of the vehicles from a monitoring apparatus as the vehicle approaches the particular section of road; and a road layout characteristic receiving module configured to receive road layout data representing the dimensions and layout of the particular section of road from the monitoring apparatus as the vehicle approaches the particular section of road. The controller may be configured to input the real-time data to the simulator as input parameters, the simulator being configured to execute a simulation in which the density of agents and/or average agent speeds in the simulation correspond to the real-time data, and the simulation environment is configured to represent the particular section of road according to the received road layout data; and the alert signal causes a warning to be generated within the vehicle.

Wherein the term approach in the above example may be taken to mean: within communication range of the monitoring apparatus and moving toward the monitoring apparatus.

In any of the implementations which utilise real-time data, the simulations using real-time data may be executed periodically or on a continual basis, and the collision rate values output recorded and those historical collision rate values used as a basis for the alert criterion. For example, if the output collision rate value using the most recent real-time data is more than a fixed percentage or a fixed (or predetermined) number of standard deviations above the mean, then the alert criterion is satisfied and the alert signal is output by the communications module. The content of a warning depends on the nature of the simulation. In an embodiment in which the subject agent represents a bicycle and the further agents motor vehicles, the warning content may be intended to remind motor vehicle driver/operators to be aware of cyclists sharing the thoroughfare and/or to provide advice regarding safe passing distances and speed.

Embodiments of another aspect include a computer-implemented collision prediction method, the method comprising, at a computing apparatus: executing an iterative agent-based simulation in which, in each iteration: a subject agent representing a travelling party travels from an end of a simulation environment representing a thoroughfare toward another end; one or more further agents representing vehicles interact with the subject agent, the simulator being configured to categorise the outcome of each interaction as either ion being complete when the subject au e outcome is categorised as a c a a collision rate value representing a number er iteration.

a c I 'i end, or is involved in an interaction of w comprising ording and outp outcome is categorised as a. col I r reaches the another ion; the method further errs in which the Embodim program which, when executed by a con ores,car emhodimen effort as the collision prediction apparatus of an Embodiments of another aspect include a computer program which,. when executed by a computing apparatus, causes the computing apparatus to become a collision prediction apparatus embodying the present invention.

Embodiments of another aspect include a computer program which, when executed by a computing apparatus, causes the computing apparatus to perform a method embodying the present invention.

The computer program may be stored on a storage medium. For c e computer program may be stored on a non-transitory storage medium.

A computing apparatus in the bo e emtbodiments includes a processor, memory, interface, and optionally a display unit and components for accepting inputs. The computing apparatus may be a single computer, or a network of computers such as a client and a server, wherein the server provides the processor and at least some of the memory and storage, and the client provides the display unit and components for accepting user inputs.

file collision prediction apparatus may also be termed a collision rate prediction apparatus or a traffic simulation platform. The apparatus or platform may be realised by hardware configured specifically for carrying out the functionality of the embodiment. The apparatus or platform may also be realised by instructions or executable program code which, when executed by a computer processing unit, cause the computer processing unit to perform the functionality attributed to the apparatus/platform. The computer processing unit may operate in collaboration with one or more of memory, storage, I/O devices, network interfaces, sensors (either via an operating system or otherwise), and other components of a computing device, in order to realise the functionality attributed to the apparatus/platform.

Althi ethods/ -a discussed separately, it sham features and consequences thereof discussed in re la sped are equally app aspects. Therefore, where a method feature is discuss d, it is aken for granted that the apparatus embodiments include a unit apparatus configured to perform that feature or provide appropriate functionality, and that programs tonfigured to cause a computing apparatus on which they are hieing executed to perform said method feature.

In any of e above aspects, the va ions features y be implemented in hardware, or as software modules running o more processors. Features of one peel may he applied to any of the other aspects.

The invention also provides a computer program or a computer program product for out any of the methods described herein, and a computer readable medium having stored thereon a program for carrying out any of the methods described herein. A computer program embodying the invention may he stored on a computer-readable medium, or it could, for example, be in the form of a signal such as a downloadable data signal provided from an Internet website, or it could be in any other form.

Preferred features of the present invention will lo be described, pu ely by way of example, with reference to the accompanying drawings, in which:-Figure 1 is a schematic diagram of a collision prediction apparatus embodying the present invention; Figure 2 illustrates an arrangement of components of an embodiment; Figure 3 illustrates another arrangement of components of an embodiment; Figure 4 is a visual representation of an agent-based model of an embodiment; Figures 5a-5c illustrate variables that describe an overtaking manoeuvre in an model of an embodiment; Figure 6 illustrates longitudinal and lateral gaps in the presence of an oncoming vehicle agent in an agent-based model of an embodiment; Figure 7 is a flow diagram describing the sequence of steps king place before an overtaking manoeuvre in an agent-based model of an embodiment; Figure 8 is a flow diagram describing the sequence of steps taking place after an overtaking ianoeuvre is initiated in an agent-based model of an embodiment; Figure 9 is a flow diagram describing the sequence of steps associated to a motor vehicle agent allocated a value of altruistic for the attitude characteristic holding back behind a cyclist agent in an agent-based model of an embodiment;; Figure 10 is a flow diagram describing the sequence of steps associated to a motor vehicle agent allocated a value of selfish for the attitude characteristic attempting to overtake a cyclist agent in an agent-based model them; an oncoming vehicle age Figure 11 is a flow diagram describing detected once an overtaking manoeuvre is under way in std model of an embodimen Figure 12 is a flow diagram describing an alternative sequence of steps taking place before an overtaking manoeuvre in an agent-based model of an embodiment; Figure 13 is a flow diagram describing the sequence of steps taking place after an overtat overtaking vehicle agent accelerates to overtake a cyclist in an agent-based model of an embodiment; and Figures 14a and 14b are exemplary visual representation of simulations performed by a simulator embodying the present invention.

Figure 1 is a schematic diagram of a collision prediction apparatus 10 embodying the present invention. The collision prediction apparatus 10 includes a controller 12 and a simulator 14. The controller 12 is optional. 15 The simulator 14 is configured to execute an iterative agent-based simulation in which, in each iteration: a subject agent representing a travelling party travels from an end of a simulation environment representing a thoroughfare toward another end. One or more further agents representing vehicles interact with the subject agent, the simulator 14 being configured to categorise the outcome of each interaction as either a collision or not a collision. Each iteration is complete when the subject agent either reaches the another end, or is involved in an interaction of which the outcome is categorised as a collision. The simulator 14 is configured to output a collision rate value representing a number of interactions in which the outcome is categorised as a collision per iteration. The collision rate value is a number, which may be accompanied by additional information, for example, statistics representing the relative speed of agents involved in a collision, or the respective speed of each agent involved in a collision.

The simulator 14 is a platform for simulating the behaviour of travelling parties on a thoroughfare. The simulator 14 comprises stored rules for determining agent behaviour, and a processing module for executing the simulation by applying the rules to the dynamic geometrical circumstances within the simulation environment. Such a processing module may be configured to execute parallel processing threads, one for each agent in the simulation.

The rules may be configurable in dependence upon attributes of the agent whose behaviour they determine. For example, the rules may be generic rules which are configurable on a per agent basis to become agent-es. Attributes of agents 12 may be provided to the 14 as arguments simulation. Alternatively, the simulator 14 itself may separate rules for allocating to agents. Alternatively, the collision prediction apparatus 10 whether via the controller 12 or by means, such as a real-time data receiving e data o rid domain, and the generic rules may he specific by using parameters from the simulator 14 may be instruetable directly horn parties external o the cc on apparatus 10.

Alter atively, the controller 12 may he configured to provide instructions to the simulator 14 in order to prompt the simulator 14 to execute a simulation. The instructions (illustrated by the arrow between the controller 12 and the simulator 14 in the figure) may include input parameters governing the form of the simulation, such as a number of iterations or a total run-time, or an instruction to run until instructed to stop. The instructions may also include input parameters for use by the simulator 14 in generating agent-specific rules for the agents in the simulation based on the stored generic rules. For example, the input parameters may specify a vehicle density, average speed of the vehicles, proportions of vehicle in cacti of a number of categories (caribicycle/HGV), and a value representing the attitude of each agent (or a proportion of agents having each value representing attitude that the simulation should fulfil). The attitude of an agent may be represented by a dynamic mathematical model that takes into account a plurality of factors influencing the attitude of a driver, the plurality of factors including one or any combination of age, type of vehicle, time of day, and traffic conditions. Input parameters may also specify features of the simulation environment, including length, width, number of lanes, width of central partition, direction of travel of each lane, existence or otherwise ol'a cycle lane, location and width of side roads, among others.

As illustrated in he collision prediction apparatus 10 may be a roadside or remote apparatus, configured to use real time data provided by a monitoring module 20 (which may be a component of the collision prediction apparatus 10 or may be a separate entity in communication with the collision prediction apparatus 10) representing the traffic on a particular section of road in the real-world domain, to predict when risk of accident is higher than normal. In such cases, the simulation environment may be configured at design time to represent the particular section of road in the real-world domain, so that certain features of the simulation environment are fixed and non-configurable. However, real-time data describing the density', profile (in terms of vehicle type), and behaviour (such as average travelling speed) may be received by the controller 12 from the monitoring module 20, and provided as input parameters to he matched or represented by the simulator 14 in an executed simulation. Alternatively, the monitoring module may be configured to provide input parameters directly to the simulator 14.

Maim 11 is configured to use the input parameters to execute a simulation having a predetermined number of iteratio tuber of iterations instructed by the controller, or having a fixed run time. The collision rate value is output by the simulator 14 after the simulation and received by the communications module l6. The communications inodule 16 is configured to determine whether or not the received collision satisfies an alert criterion. As an example of such an alert criterion, the communications module tore the collision rate values output on a rolling basis to cover a fixed length of time preceding the If the most recently output: collision rate value is more than a predetermined (for example, I, 2, number or standard deviations above the mean of the stored collision rate values, then the alert criterion Ii the alert criterion is satisfied, then an alert signal is output. This may have one of a number of effects, depending on the particular implementation.

In the example of Figure 2, the communications module is connected to configurable signage 30 at the roadside in the real world domain. The configurable signage, upon receipt of the alert signal, may he configured to display a predetermined message, such as a warning that collisions are likely, or a recommended speed limit (that is lower than the legal maximum at the particular section of mad). The configurable signage 30 may be a component of the collision prediction apparatus 10 or may be a separate entity in communication with the collision prediction apparatus 10. Other effects caused by the alert signal may be to alert emergency services or an authority responsible for highway management that there is an enhanced risk of collisions at the particular section of road. Alternatively the alert signal may be a broadcast message over radio, Bluetooth, or some other wireless medium, that can be received by a ear or by a mobile terminal within a car and a warning message output audibly.

In the example of Figure 3, the collision prediction apparatus 10 is a component of a motor vehicle 40. The collision prediction apparatus 10 in Figure 3 is the same as that of Figure 2, with the exception of the connection to the configurable signage. In Figure 3, an alert signal output by the communications module produces a warning, either audibly or visually on the dashboard or by some other means, to the operator of the motor vehicle 40, which may be a car, that there is an enhanced likelihood of collisions occurring at the current section of road along which the motor vehicle is currently travelling. The monitoring module 20 is a roadside apparatus configured to output data over radio, Bluetooth, will, or in some other wireless means, that is received by the collision prediction apparatus 10 onboard the vehicle 40, and used as the basis of input parameters to the simulator 14. The simulator 14 therefore performs a simulation representing the current (or current at the most recent gathering of data by the monitoring module 20) real-world situation on the particular section of road.

A detailed description of a simulator 14 will now be presented. The simulator 14 is a traffic simulation platform based on an agent-based model (AIM) re tutor vehicles and a cyclist as agents moving along a road. The cyclist is exemplary of a subject agent and the motor vehicles arc exemplar/ of further agents. As an illustrative example, we will consider the situation depicted in Fig. 4: two agents 141 and 143 representing motor vehicles move along a rectilinear stretch of road, one going northwards and the other southwards, while another agent representing a cyclist 142 is moving northwards; in this figure the north is indicated by the N at the top. This scenario is used to exemplify features of the traffic simulation platform.

Agents move withi ion nt that represents a road with eometry and layout; these can be changed by a controller or instructing party providing arguments (input parameters) to the traffic simulation platform. For example, features of the geometry and layout may be modified to investigate the impact such modifications have on the synthetic statistics generated by the traffic simulation platform; and therefore to aid decision making processes regarding whether and how to modify the represented road in the real world domain. In the example shown in Fig. 4 the dimensions of this environment correspond to an actual stretch of road 100 metres long (A4007, between Iver Heath and St. John's Road, Greater London), with a width of 7.25 metres and a central separation 1.35 metres wide. The visual interface shown in Fig. 4 is an optional mechanism for visualising the simulator 14 dynamics: since the aim of the simulator 14 is to generate synthetics statistics the visual interface may be used only during an initiation phase to view the simulation environment, or not at all.

Agent dimensions dimernsio ns correspond to those of actual vehicles: in this example motor vehicle dimensions correspond to a medit -sized hatchback car, while bicycle dimensions are assumed to be 1.8 metres long with handlebars 0.65 metres wide. 01 course, a plurality of different combinations of characteristics (dimensions, acceleration capabilities, manoeuvrability, etc) can be made available for agents, with the simulator 14 selecting a combination as a new agent enters the simulation environment. Visibility distances for each and every agent can be specified to account for time of the day, weather conditions, and the use or not of a high-visibility outfit by the cyclist. Again, at each iteration and/or as each new cyclist agent enters the simulation environment, the simulator 14 can allocate characteristics such as whether or not high-visibility clothing is being worn, whether or not a helmet is being worn, speed of travel etc. Characteristics attributed to agents may he selected based on parameters input to the simulator 14, such as a prescribed proportion of agents in the simulation which are to be attributed each of the characteristics. For example, an input parameter may be that 80% of cyclists wear helmets, so that the simulator can ensure that the prescribed proportion is adhered to over the course of the simulation. A random element may be included in the selection of characteristics prevent patterns of different characteristics being established relative to one another. The speeds at which vehicles move can also be attributed to the agents: in this particular u ple ears move at a nominal speed of 40 mph, while the cyclist moves at a nominal speed of 11 mph. Of course, acceleration and deceleration at take place during manoeuvres. For situp icity'ssake we have assumed periodic boundary conditions particular ample; that is, once an agent moves beyond the northernmost or southernmost edge of the road it reappears at the opposite one Rach time the cyclist agent 142 reaches the northernmost edge, a anon is deemed complete. As motor vehicle agents and the cyclist progress along the simulation environment, the northbound vehicle agent 141 may have to perform an overtake manoeuvre in order to pass the cyclist; this manoeuvre can be affected by several factors as described in the next paragraphs.

The northbound motor vehicle agent 141 may also be referred to as the potential overtaking agent, or, for simplicity, the overtaking agent. It is noted that the overtaking agent 141 does not overtake in every scenario, but is travelling in the same direction as the subject agent. The subject agent 142 in this example is a cyclist agent. The southbound motor vehicle agent 142 may also he referred to as the oncoming vehicle agent 143.

Sample input parameters o the simulation illustrated in Figure 4 arc as fol 1div=0.25m ROAD DIMENSIONS: Road length -100.00m Road width = 7.25m Separator width -1.35m Lane width =2.95m Cycle lane width = 0.00m CAR CHARACTERISTICS: Car length = 3.97m Car width = 1.68m Car visibility= 1.00m DRIVER CHARACTERISTICS: Selfish driver B ICATCL A rS: Dike length = 1,80w Bike width = 0.65w Distance from kerb = 0. 75w Bike visibility= 25m

RIDER CHARACTER

Male rider I Ielinet Some or all of the above i t parameters may cepted by a simulator 14. Among those accepted, some or all may have default values which arc used in the simulation if no value is provided at initiation/instruction of the simulation. The "Car" and "Driver" characteristics relate to the potential overtaking agent 141, which in this simulation is a car. The subject agent 142 is a cyclist agent represented by the "Bicycle" and "Rider" characteristics. The simulation environment represents a road. In the example, the characteristics have been given as true for all cases. Alternatively, a proportion could be given, such as selfish driver = 0.55 (and hence altruistic driver -0.45), Male Rider = 0.5 (and hence female rider = 0.5), so that each time an iteration is complete, or each time an agent leaves the simulation environment and is eintroduced at the other end, the simulator 14 creates a new agent having characteristics selected to achieve the specified proportion over the course of, for example, each 100 iterations.

abort variables at t--"I Os are as follows:

SPEEDS

Northbound car -40.1/0 mph Southbound ear = 40.00 mph Cyclist= 11.00 mph

DISTANCES

carN-bicycle = 0.00 cars-bicycle = 0.00m ears-carN = 0.00 m dpassX = 0.00 m [Massif VERTAKING EVENTS: Collision threshold --. 0.05 m Overtaking events = 0 Accidents -0 = 0.00 MEiSSAGES: (BLANK) After time t=0.00s the distances will be initiated (via measurements performed by the simulator 14). The speeds arc the nominal speeds of the agents, which in the present example are 40.00mph for the "cars' potential overtaking agent 141 and the oncoming vehicle agent 143) and 11.00mph for the "bicycle" (cyclist agent 142).

s 5a-5c show some exemplary variables that describe an overtaking manoeuvre in the traffic simulation platform: Yp"", is the longitudinal distance at which the overtaking manoeuvre is initiated (Fig 5a), Xp", is the lateral distance between the cyclist agent 142 and the northbound motor vehicle agent (Fig. 5b) as the overtaking progresses, and Y1,a"2 is the longitudinal distance at which the overtaking manoeuvre is about to be completed (Fig. 5c). Once a cyclist agent 142 has been seen by a driver (that is, once the distance between the cyclist agent 142 and the approaching northbound motor vehicle agent is less than or equal than the visibility distance specified for the cyclist agent 142 given the conditions affecting visibility attributed to the simulation/iteration such as weather, time of day, and type (visibility.) of outfit worn by cyclist) the northbound motor vehicle agent 143 performs an estimation of these distances. This estimation is exemplary of a geometrical assessment performed by the agent. Of course, agents are modelled entities in the traffic simulation platform, and hence assessments or movements performed by agents are in fact decisions made by the simulator 14. Examples of the mariner in which any of the geometrical assessments made by the agents can be performed in the traffic simulation platform include through a deterministic procedure or through generation from a random distribution: for example, \Fitts,' and Yp""2 could be assumed as equal to one car length, while Xpas can be generated using measured statistics on overtaking distances that take into account the cyclist's gender, use of helmet, and distance from the kerb. Such deterministic values may be available from monitoring actual sections of road in the real-world domain and gathering statistics, and/or from publications such as Accident Analysis and Prevention 39 (2007) 417-423. in this example, there are two other variables that determine the behaviour of the overtaking motor vehicle agent if an oncoming is detected during the overtaking manoeuvre: the longitudinal and lateral gaps be cyclist agent 142 and the oncoming agent 143, as shown in Fig. 6. Both gaps can be assessed by the overtaking agent using standard equations of motion, tional addition (or inclusion in some other er or factor to account for driver tnisealeuIation in the real-world domain.

noted that longitudinal is taken to mean along the path defined by the road or thoroughfare,and lateral to mean across the path defined by the road or thoroughfare.

The sequence of steps followed by a motor vehicle agent before perforating an overtaking manoeuvre detailed in the flow chart shown in Fig 7. The flow starts at step SOO; at step S401 the overtaking agent verifies if a cyclist agent 142 has been seen (that is, if the distance between the overtaking agent 141 and cyclist agent 142 is less or equal than the visibility distance specified for the cyclist agent 142 in the simulation). If this is not the ease the car agent continues its advance at step 5402, but if a cyclist agent 142 has been seen the lateral and longitudinal passing distances Xpa", Ypaso and Yp""2 are assessed by the overtaking agent 141 at step 5403. At step S404 the overtaking agent 141 verifies if it is behind the cyclist agent 142 and if the distance between itself and the cyclist agent 142 is less than or equal to the longitudinal passing distance Yo"1: if this is not the case the ear agent advances at step S405, otherwise the flow continues to step 5406 where the overtaking agent 141 verifies whether or not there is an oncoming vehicle agent 143 (that is, if the distance between the overtaking agent 141 and the oncoming vehicle agent 143 is less than the visibility distance specified for the latter). If there is no oncoming vehicle agent 143 visible the flow continues to step 5408 where the overtaking agent 141 initiates the overtaking manoeuvre by moving a distance Xp", towards the outside lane, with the flow then continuing to step 5500. lf there is an oncoming vehicle agent 143 the overtaking agent 141 assesses (or estimates) the longitudinal gap between the cyclist agent 142 and the oncoming vehicle agent 143 at step 5407. At step 5409 the overtaking agent 141 verifies if this longitudinal gap is big enough to start the overtaking manoeuvre; if this is the ease the flow proceeds to step 5408 where the overtaking manoeuvre is initiated, with the flow then continuing to step 5500.

There are two possible actions if the longitudinal gap between the overtaking agent 141 and the oncoming vehicle agent 143 is assessed as being not big enough after step 5409; each can be associated to an attitude characteristic attributed to the overtaking agent 141 by the simulator 14 as an adjustable parameter: altruistic attitude or selfish attitude. The attitude characteristic is attributed to each motor vehicle agent either at the start of a simulation or at the start of an iteration within a simulation. It may be that a controller or other entity instructing the simulation specifies a proportion of motor vehicle agents that should have each value (altruistic/selfish), and the simulator 14 is configured to match the specified proportion over the course of the simulation. Of course, n simply binary values, there may be more values on a scale rep esenting the attitude eharac mplc use case and to illustrate the principle of the attitude characteristic, tic for the attitude characteristic (that is to say, representing a motor vehicle having hold hack behind the a distance Y,,.r until the oncoming vehicle 43 has passed, whereas an agent having a value of selfish for the attitude characteristic (that is to say,representing a motor vehicle having a value of selfish for the a titude characteristic) will te the lateral gap between the cyclist agent 142 and the oncoming vehicle agent 143, and if this gap is large enough, the overtaking agent 141 will htitiate the overtake manoeuvre regardless of oncoming traffic; otherwise it will be forced to hold back behind the cyclist agent 112 at a distance Yp"", until the oncoming vehicle agent 143 is left behind. These two possibilities are contemplated at step S1[0: if the overtaking agent 141 has an altruist attitude characteristic (which may be referred to as having a motor vehicle agent allocated a value of altruistic for the attitude characteristic) the flow is diverted to step 5600, otherwise the flow continues at step 5700.

The flow diagram in Fig. 8 describes the geometrical assessments and consequential movements that take place after the overtaking manoeuvre has been initiated at step 5408. After step 5500 the overtaking agent 14I verities if there is an oncoming vehicle agent [43 at step 5501: if that is not the case the flow continues to step 5502, where the overtaking agent 141 assesses whether or not it is ahead of the cyclist agent 142 and whether or not the distance between itself and the cyclist agent 142 is equal or greater than Ypass2. If the overtaking agent 141 assess that it is ahead of the cyclist agent 142 by a distance equal to or greater than Ypass2, the overtaking agent 111 completes the overtaking manoeuvre by moving away from the outside lane by a distance)(pass at step 5504, with the flow then being transferred to step 5400. If an oncoming vehicle agent [43 is detected while the overtaking manoeuvre is underway at step 5501 there are two possible actions that could take place at step 5800: if the overtaking agent 141 has an altruist value of the attitude characteristic it will abort the overtaking manoeuvre by moving back into the inside lane and remaining a fixed distance behind the cyclist agent 142 until the oncoming vehicle agent 143 has passed; if on the contrary the overtaking agent 141 has a selfish value of the attitude characteristic it will continue the overtaking manoeuvre if there is a large enough lateral gap between the cyclist agent 142 and the oncoming vehicle agent 143, otherwise the overtaking agent 141 will be forced to abort the overtaking manoeuvre and hold back behind the cyclist agent 142 at a safe distance until the oncoming vehicle agent 143 has passed.

The distance at which the car agent will hold back behind the cyclist went 142 can he calculated in a deterministic manner or can be generated from measurement statistics.

Figure 9 describes the hellcat 24 motor e agent allocated a value of altruisti characteristic at step 5600 when the overtaking el not started, an oncoming vehicle agent 143 has been detected, and the longitudinal gap between 142 and the oncoming vehicle agent not big enough for a safe overtaking manoeuvre; in this case the car agent holds back behind the cyclist agent 142 at a distance Y"""i. The flow starts at S60 i the car agent verifies if its speed matches the speed of the cyclist agent 142, and if that not e case the speed of the car agent is decreased by a certain amount at step 5602, with the flow then refuta p q u ired braking acceleration can be calculated using standard equations of motion. Once the car agent atches the speed of the cyclist agent 142 the flow continues at step S603, where the car agent holds back behind the cyclist agent 142 until the oncoming vehicle agent 143 is behind the car agent; at that moment the overtaking manoeuvre is initiated at step 5604 when the car agent moves a distance X. towards the outside lane; the flow then continues at step S500.

Figure 10 describes the behaviour of an overtaking agent 141 allocated a value of selfish for the attitude characteristic at step 5700 when the overtaking manoeuvre has not started, an oncoming vehicle agent 143 has been detected and the longitudinal gap between the cyclist agent 142 and the oncoming vehicle agent 143 is assessed as not being big enough for a safe overtaking manoeuvre; in this case the overtaking agent 141 estimates the lateral gap between the cyclist agent 142 and the oncoming vehicle agent 143 in order to attempt overtaking the cyclist in spite of an oncoming vehicle agent 143. After the flow starts at step 5700 the overtaking agent 111 estimates (that is, makes an assessment ot) the lateral gap between the cyclist agent 142 and the oncoming vehicle agent 143 at step 5701; if this gap is assessed as being greater than the width of the overtaking agent 141 the flow continues at step 5703, otherwise the overtaking agent 141 holds back behind the cyclist agent 142 at step 5600. At step 5703 the overtaking agent 141 makes an assessment to verify wvhether or not its course must be corrected (i.e. is any lateral or longitudinal acceleration required) in order to pass through the gap between the cyclist agent 142 and the oncoming car agent, if that is the case the value of X,,"" is corrected at step 5704. The flow then proceeds to step S705 where the overtaking agent 141 vitiates the overtaking manoeuvre by moving towards the outside lane a distance Xpass; once the overtaking manoeuvre is underway the flow continues at step 5500.

Figure II describes the sequence of assessments and consequential actions/movements executed by the traffic simulation platform when an oncoming vehicle is detected once an overtaking manoeuvre is under way: in such a case a motor vehicle agent allocated a value of altruistic for the attitude characteristic will immediately abort the overtaking manoeuvre, whilst a motor vehicle agent allocated a value of selfish for the attitude characteristic will try to complete the manoeuvre if there is a big enough gap between the cyclist agent 142 and the Oncomutg vehicle agent 143, othe wise aborting the manoeuvre. After the flow starts at step 8800 the actions take agent will depend on its driver type at step 880 4 If the agent;ent allocated a value of altruistic for the attitude characteristic, incl the cyclist agent 142 at step 802; this safe distance could performing the overtake is a motor vein the agent will verify if it is at a safe distal be determined in a deterministic fashion (e.g. safe distance equal to Yq3"1) or generated as a random selection from a distribution that accounts for observed statistics longitudinal separation does a motor vehicle. allow when pulling in behind a cyclist). The overtaking agent 141 will brake at step 8803 (that is to say, will reduce speed in the longitudinal direction) until the requirement of step 8802 is satisfied; once this happens the overtaking manoeuvre is effectively undone at step 5804 when the overtaking agent 141 moves a distance Xpa" away from the outside lane. The overtaking agent 141 then matches the speed (in the longitudinal direction of travel) of the cyclist at step 8805 and holds back behind it at step 8806. Once the oncoming vehicle is behind the overtaking agent 141 at step 5806 the overtaking agent 141 initiates the overtaking manoeuvre again at step 5807 by moving a distance X"an towards the outside lane and by accelerating to reach the overtaking agent 141's nominal speed at step 5808. The flow then proceeds to step 5500 to eventually complete the overtaking manoeuvre.

If the overtaking agent 141 is allocated a value of selfish ter the attitude characteristic, the overtaking agent. 141 will estimate the lateral gap between the cyclist agent 142 and the oncoming vehicle agent 143 at step 5809. If this lateral gap is not big enough at step 8810 the overtaking manoeuvre is aborted and the flow proceeds to step 8802; if that is not the case the overtaking manoeuvre continues at step 8811, where the overtaking agent 141 determines if a course correction is necessary in order to drive through the lateral gap; if this is necessary-the required value of is is calculated and applied at step 8812. The flow then proceeds to step 8813, where the overtaking agent 141 verifies if it is at a distance Yp",2 ahead of the cyclist; if that is not the case the overtaking agent 141 is not yet ready to complete the overtaking manoeuvre so it continues its advance at step S814. If another oncoming vehicle appears while the overtaking agent 141 is advancing in the outside position at step 8815 the flow proceeds to step 5809 in order to estimate the lateral gap between the cyclist and oncoming vehicle agents 143; if there is not a new oncoming vehicle the flow proceeds to step 8813. Once the condition in step 8813 is satisfied the overtaking manoeuvre is completed at step S816 by moving the overtaking agent 141 a distance Xp", away from the outside lane; the flow then proceeds to step S400. The lateral gap that is considered big enough at step 8810 may vary from one incident to another. For example., by monitoring traffic in the real world domain a distribution may be established of lateral gaps between an oncoming vehicle and a cyclist being overtaken that an overtaking vehicle will attempt to fit through in order to execute an overtaking manoeuvre. Of course, it tnay be the case therefore that a lateral gap which is not actually big enough for the overtaking motor vehicle agent to fit through will be deemed big step S810, because the distribution will include some f overtaking vehicles have attempted to fit threw ich are too small. In such cases, collisions In the preceding description it has been assumed that the overtakingagent 141 does not accelerate order o overtake a cyclist agent 112; however overtaking motor vehicle agents could also increase their speed in order to drive through the longitudinal gap between a cyclist agent 142 and an oncoming vehicle agent 143. The forthcoming description and accompanying figures describe an alternative or additional elements of the modelling that incorporate acceleration in an overtake, Figure 12 shows a modification of the flow diagram of Figure 7: the flow starts at step S900 and at step 5901 the overtaking agent 141 determines it a cyclist agent 142 has been detected (distance between overtaking agent 141 and cyclist agent 142 less than or equal to the visibility distance specified for the cyclist agent 142). If that is not the case, the overtaking agent 141 continues its advance at step S902. If a cyclist agent 142 is detected the lateral and longitudinal passing distances Xp"" Yp",, and Y,,a"2 are determined by the overtaking agent 141 at step S903. At step 5904 the overtaking agent 141 verifies if it is behind the cyclist and if the distance between itself and the cyclist agent 142 is less than or equal to the longitudinal passing distance Ypassi: if this is not the case the overtaking agent 141 advances at step 5905, otherwise the flow continues at step S906 where the overtaking agent 141 verifies if there is an oncoming vehicle agent 143 (distance between overtaking agent 141 and oncoming vehicle agent 143 less than the visibility distance specified for the latter). If there is no oncoming vehicle agent 143 the flow continues to step S907 where the overtaking agent 141 initiates the overtaking manoeuvre by moving a distance X. towards the outside lane, with the flow then continuing to step S500.

Of course, step S904 is an assessment performed by the overtaking agent 141 and therefore may incorporate a random factor to simulate inaccurate geometrical assessments made by drivers in the real world domain.

For example, a random factor (which may be positive or negative) may be added to the actual distance between the overtaking agent 141 and the cyclist agent 142 in the simulation environment, and the result is the distance compared with Ypassl at step 5904. The incorporation of random factors in such a manner may be incorporated into all verifications, estimations, assessments, and/or movements performed by the agents.

The magnitude and distribution of the random factor (i.e. the potential extent and the likelihood of each value within the potential extent) may be determined as a global variable of the simulation, such as plus/minus 10%, or may be determined by observation of real-world behaviour.

If an oncomingvehicle agent 143 is detectet and oncotnin acceleration 0 between the cyclist and oncoming S906 the long di d at step S908; the overt/ e overtaking agent 141 ti gent 142 agent 141 then estimates the rough the longitudinal gap nts 143 at step 5909. The flow then continues at step 8910 where the overtaking agent 141 determines if it the acceleration required to drive through the longitudinal gap. Estimates at steps 8908, 8909 and 5910 can be obtained using standard equations of motion plus a random additive factor to account for driver miscalculation, as well as data such as the 0 to 100 km/h acceleration time for the particular vehicle that is being modelled by the overtaking agent 141. If the overtaking agent 141 can achieve the required acceleration the flow proceeds to step 81000, otherwise the flow continues to step 891: if the overtaking agent 141 is a motor vehicle agent allocated a value of altruistic for the attitude characteristic the flow is diverted to step 8600 to abort the overtaking manoeuvre, otherwise the flow continues at step 5700.

Figure 13 shows the sequence of steps required when an overtaking agent 141 accelerates to drive through the longitudinal gap between a cyclist agent 142 and an oncoming vehicle agent 143. The flow starts at step Si 000, and then at step 81001 the overtaking manoeuvre is initiated with the overtaking agent 141 movin<:, towards the outside lane a distance X,,",. At step 51002 the overtaking agent 141 verifies if it has acquired the speed required to complete the overtaking manoeuvre; if that is not the case the speed of the overtaking agent 141 is increased at step 51003 by an pm-established amount previously calculated through standard equations of motion plus data such as the 0 to 100 km/h acceleration time for the particular vehicle that is being modelled by the overtaking agent 141. The flow then proceeds to step 51004 where the overtaking agent 141 continues its advance. If the overtaking agent 141 has acquired the necessary speed at step S1002 the flow continues to step 51005 to verify if the overtaking agent 141 is ready to complete the overtaking manoeuvre; if that is the case the overtaking agent 141 moves away from the outside lane a distance X,"" at step 51 006. If at step SI005 the overtaking agent 141 is not ready to complete the overtaking manoeuvre the flow returns to step 51004 to continue its advance. After step SI006 has been executed the overtaking agent 141 decelerates to acquire its nominal speed at steps 51007 and 51008, with the flow then proceeding to step 5900.

The nominal speed of each agent is a variable attributed to the agent during an initiation phase of the simulation or of each iteration and is a default speed at which the agent travels along the thoroughfare of the simulation environment when not involved in manoeuvres which necessitate longitudinal acceleration or deceleration that is, changing the speed.

additional mod itic 28 elements of traffic behavic be introduced to the scenario described so far to simulate additional hence low processing overhead) o modifications include introducing occasional swerving of a cyclist ager e at the same time retaining the simplicity (and at describe the behaviour of the agents. Examples of auctions to model parked or broken down cars, representing the s path to avoid potholes or drain grids ((i.e. this can he a random lateral movement of the cyclist agent 142 that occurs with a predetermined probability per unit distance travelled and has a randomly selected lateral extent), allowing variations in the nominal speed of the vehicles, representing the die effectof traffic lights, etc. In the traffic simulation platform, collision ae recorded whenever agents move such that they become spatially coincident.. This may he due to inaccurate geometrical assessments performed by the overtaking agent 141, and attempting manoeuvres for which there is not sufficient space. Agents may be modelled in such a way that they take action to avert collisions when the course of other agents is such that a collision will occur without any such action. flowever, even if agents arc not modelled to take specific action to avert collisions, the collision rate statistics will be meaningful, since they can represent a rate of manoeuvres occurring which would result in a collision without specific avoidance actions. Based on an assumption of a fixed percentage of such manoeuvres actually, resulting in a collision in the real-world domain, the collision rate statistics for different road layouts or other variable parameters can be compared to reveal which parameters generate the highest probability of collisions occurring.

Collision rate statistics can be generated by the simulator 14 for a given set of input parameters and recording the average number of accidents that occur per overtaking event or per iteration (wherein an iteration may be an attempted journey along, the simulation environment by a subject agent such as a cyclist agent 142). There are various possible ways in which the occurrence of a collision could be accounted for: for instance, a distance-speed threshold could be defined such that a collision is recorded whenever the distance between a cyclist agent 142 and a motor vehicle agent is less than a distance certain threshold while the motor vehicle agent's speed exceeds a speed threshold (or each time the distance-speed threshold is satisfied a probability distribution could be used to determine whether or not a collision is deemed to have occurred). The influence of factors such as road dimensions and layout, vehicle speeds, driver and rider attitudes, visibilities, etc. on the likelihood of collisions can thus be accounted for in a quantitative way, thus providing a means by which to measure the cost/benefit ratio of measures aimed at improving road safety.

Figures 14a and 14b shows an example of the output of a visual interface representing the simulation environment and displaying input parameters and other variables to a user. In the particular example of central separation, which has been reduced in Figure 4 For a isibilities of 100 and 25 metres respect( verage distance of 0.75 metres from t 14b shows the same road but wit c Figure 14a sltuws the original Ia e input parameters to the simulatio Idiv=0.2Sm ROAD DIMENSIONS: Road length -100.00in Road width = 7.25m Separator width = I.35m Lane width -295m Cycle lane width -0.00m CAR CHARACTERISTICS: Car length = 3.97w Car width I.68m visibility-100m DRIVER CI LARAC ITKISTICS: Selfish driver BICYCLE CHARACTERISTICS: Bike length = 1.80m Bike width = 0.65m Distance from kerb = 0. 75m Bike visibility= 25m RIDER CHARACTERISTICS: Male rider I lelmet d in mure I La are as follows: And in-simulation I I Figure [4a are as follows:

SPEEDS

Northbound car = 40.00 mph Southbound car= 40.00 mph I I,00 mph

DISTANCES

Cyclist not seen yet Oncoming car not seen yet carS-earN = -1.82 in dpassX 0.00 in dpassY -0.00 in OVERTAKING EVENTS: Collision threshold 0.30 m Overtaking events = 300 Accidents = 6 P(A)= 0.02 MESSAGE'S. (BLANK) Figure 14b is shown at t = 2510.8593s, so is beyond the first iteration (in fact, approximately 300 rations have been performed in the simulation). The input parameters have been adjusted from those used in Figure I 4a to investigate the effect of a narrower road with a narrower central separator. Sample input parameters to Figure 1.4h: t-0.25rn ROAD DIMENSIONS: Road length = 100.00 in Road width = 6.10 m Separator width = 0.20 m Lane width = 2.95 ni kith -0.0 RACTERISTICS: Car length = 3.97m Car width 1.68m Car visibility= I00ni DRIVER. CHAR' TER!SIM'S: Selfish driver BICYCLE CHARACTER!STK:Si Bike length = .80 Bike width -0.65 m Distance from kerb = 0. 75 in Bike visibility-25 in RIDER CI RACI" ' Male rider Let And on variables at t=25 10.8593s are as follows:

SPEEDS

Northbound car = 40.00 mph Southbound car =40.00 mph Cyclist = 11.00 mph

DISTANCES

Cyclist not seen yet Oncoming car not seen yet carS-carN = -14.55 m dpassX = 0.00 in dpassY = 0.00 m OVERTAKING EVENTS: Collision threshold -0.30 m Overtaking events = 322 Accidents = 69 Al 0.21 MESSAGES (13 NK) 0 ation shows that eliminating the central separation increases the likelihood of a eye! ng fated collision from 2% to over 20%.

Claims

CI MS1. A coil the simulator el iteration: a subject apparatus, the apparatus d to execute an iterative agent-based simulation in resentin cluing is fioan an end of a simulation environmei representing a thoroughfare toward another end; subject agent, the simulator bei one or more further agents representing vehicles interact configured to categorise the outcome of each interaction as either a collision or not a collision; the iteration being complete when the subject agent either reaches the another end, or is nteraction of which the outcome is categorised as a collision; the simulator being configured to record and output a collision rate value presen g a number of nteractions in which the outcome is categorised as a collision per iteration.
2. A collision prediction apparatus according to claim 1, wherein, in each iteration, the one or more further agents interacting with the subject agent includes a potential overtaking agent from among the further agents being positioned further away from the end of the simulation environment toward which the subject agent is moving, the potential overtaking agent moving toward the same end of the simulation environment.as the suhject agent at a velocity which, if maintained, would result in the potential overtaking agent reaching the end of the simulation environment before the subject agent.
3. A collision prediction apparatus according to claim 2, wherein the movements of each of the agents within the simulation environment are determined by a set of Hiles attributed to the agent by the simulator during an initiation phase of each iteration, the set of rules governing geometrical assessments used by the agent during interactions with other agents, and consequential movements based on the geometrical assessments.
4. A collision prediction apparatus according to claim 3, wherein the geometrical assessments used the one or more agents when overtaking the subject agent include one or more of the following: longitudinal and/or lateral distance between the overtaking agent nd the subject agent; longitudinal and/or lateral distance between the overtaking agent and any oncoming agent.A cotlisio pparatus according to claim 4, giber each of the geometrical assessments performed by the agent put to one or more rules governing which geometrical assessment -ulated by I g the true/actual value of the respective distance hick is used by the ake, and wherein the 6. A collision prediction apparatus according to any of claims 3 to 5, wherein the one or more rules governing which consequential movement to make include t what speeds, and by what lateral and longitudinal margin, for the potential laking age subject agent.7. A collision prediction apparatus according to claim 6, wherein the subject agent and/or one of the one or more further agents are allocated a value representing attitude during an initiation phase of each iteration, and the outcome of applying the one or more rules governing which consequential movement to make is at least partially dependent upon the value representing attitude; and/or wherein the attitude of an agent may be represented by a dynamic mathematical model that takes into account a plurality of lectors influencing the attitude of a driver, the plurality of lectors including one or any combination of age, type of vehicle, time of day, and traffic conditions.A collision prediction apparatus according to one or more of the preceding claims, further comprising: a controller configured to specify a set of input parameters and to instruct the simulator to execute a number of iterations with the specified input parameters, wherein the input parameters include one or more of the following: the dimensions and/or layout of the road represented by the simulation environment; the visibility conditions in the simulation environment; the density of agents in the simulation environment and/or the average speed of le agents; the visibility of each agent in the simulation environment; a proportion of further agents belonging to each of a predetermined set of categories of agent, including bicycle, ear, TIGV, and/or motorcycle; a. category of agent to which the subject agent belongs from among a predetermined set of categories of agent, including bicycle, car, HGV, and/or motorcycle; T the subject c belongs to the category of bicycle, other characteristics incitrding gender of rider, and whether or not the c of the agent, and/or distance between the at wherein the 'but to each agent it ase is configurable cording to the input parameters.A collision prediction apparatus at:torch din the apparatus further comprises a road layout selection module, being configured to generate the simulation environment in a plurality of versions, each version differing from one another in dimensions and/or layout, and to output the generated simulation environments to the simulator; the simulator being configured to execute a predetermined number of iterations for each version of the simulation environment, with any other input parameters being the same for each version; the road layout selection module being further configured to receive the value output by the simulator for each version of the simulation enviromnent, and to identify and output the version of the simulation environment having the lowest collision rate as a suggested road layout.10. A collision prediction apparatus according to any of the preceding claims,further comprising: a graphical representation generator configured to generate and output a graphical representation of the simulation environment and the agents in the simulation environment.II. A collision prediction apparatus according to claim 1, further comprising: a communications module configured to generate and output an alert signal when the value output by the simulator satisfies an alert criterion.12. A collision prediction apparatus according to claim I 1, wherein: the simulation environment is based on a particular section of road, and real-titre data representing the density of vehicles on the particular section of road and/or the average speed of the vehicles is monitored by a monitoring module, and the controller is configured to input the real-time data to the simulator as input parameters, the simulator being configured to execute a simulation in which the density of agents and/or average agent speeds in the simulation correspond to the real-time data, and the simulation environment represents the particular section of road; and the alert signal causes configurable s gnage located at the particular ection of road to display a warning.apparatus according 11, where( apparatus is a component vehicle; rec speed of the v and on prediction e data representin hicies front a mot ratr.rs further comprises a real-time data reeeiving module con y of vehicles on a particular section of road and/or the average he,ehicle)(mooches the particular section of road; C actis inn. an a road layout ving module configured to receive road layout data representing the dimensions and layout of the particular section of road from the monitoring apparatus as the vehicle approaches the particular section of road; the controller is configured to input the real-time data to the simulator as input parameters, the simulator being configured to execute a simulation in which the density of agents and/or average agent speeds in the simulation correspond to the real-time data, and the simulation environment is configured to represent the particular section of road according to the received road layout data; and the alert signal causes a warning to be generated within the vehicle.14. A computer-implemented collision prediction method, the method comprising, at a computing apparatus: executing an iterative agent-based simulation in which, in each iteration: a subject agent representing a travelling party travels front an end of a simulation environment representing a thoroughfare toward another end; one or more further agents representing vehicles interact with the subject agent,the simulator being conf ured to categorise the outcome of each interaction as either a collision or not a collision; the iteration being complete when the subject agent either reaches the another end, or is involved in an interaction of which the outcome is categorised as a collision; the method further comprising recording and outputting a collision rate value representing a. number of interactions in which the outcome is categorised as a collision per iteration.15. A computer program which, when executed by a computing apparatus, causes the computing apparatus to function as the collision prediction apparatus as claimed in any of claims I to 13.Amendments to the claims have been made as followsCLAIMSA collision prediction apparatus, the apparatus comprising a simulator; the simulator being configured to execute an iterative agent-based simulation in which, in each iteration: a subject agent representing a travelling party travels from an end of a simulation environment representing a thoroughfare toward another end; one or more further agents representing vehicles interact with the subject agent, the simulator being configured to categorise the outcome of each interaction as either a collision or not a collision; the iteration being complete when the subject agent either reaches the another end, or is involved in an interaction of which the outcome is categorised as a collision; the simulator being configured to record and output a collision rate value representing a number of interactions in which the outcome is categorised as a collision per iteration; the collision prediction apparatus thither comprising: a communications module configured to generate and output an alert signal when the collision rate value output by the simulator satisfies an alert criterion, wherein: (r) the simulation environment is based on a particular section of road, and real-time data representing the density of vehicles on the particular section of road and/or the average speed of the vehicles is monitored by a monitoring module, and the controller is configured to input the real-time data to the simulator as input parameters, the simulator being configured to execute a simulation in which the density of agents and/or CO average agent speeds in the simulation correspond to the real-time data, and the simulation environment represents the particular section of road; and the alert signal causes configurable signage located at the particular section of road to display a warning.2. A collision prediction apparatus according to claim 1, wherein, in each iteration, the one or more further agents interacting with the subject agent includes a potential overtaking agent from among the thriller agents being positioned further away from the end of the simulation environment toward which the subject agent is moving, the potential overtaking agent moving toward the same end of the simulation environment as the subject agent at a velocity which, if maintained, would result in the potential overtaking agent reaching the end of the simulation environment before the subject agent.3. A collision prediction apparatus according to claim 2, wherein the movements of each of the agents within the simulation environment are determined by a set of rules attributed to the agent by the simulator during an initiation phase of each iteration, the set of rules governing geometrical assessments used by the agent during interactions th other agents, and consequential movements based on the geometrical assessments.4. A collision prediction apparatus according to claim 3, wherein the geometrical assessments used the one or more agents when overtaking the subject agent include one or more of the following: longitudinal and/or lateral distance between the overtaking agent and the subject agent; longitudinal and/or lateral distance between the overtaking agent and any oncoming agent.
5. A collision prediction apparatus according to claim 4, wherein each of the geometrical assessments performed by the agent gives rise to a result which is used by the agent as an input to one or more rules governing which consequential movement to make, and wherein the result of the geometrical assessment is calculated by randomly selecting one value from a range of values including the true/actual value of the respective distance in the simulation.
6. A collision prediction apparatus according to any of claims 3 to 5, wherein the one or more rules governing which consequential movement to make include whether, when, at what speeds, and by what lateral and longitudinal margin, for the potential overtaking agent to overtake the subject O agent.dS
7. A collision prediction apparatus according to claim 6, wherein the subject agent and/or one of the one or more further agents are allocated a value representing attitude during an initiation phase of each iteration, and the outcome of applying the one or more rules governing which consequential movement to make is at least partially dependent upon the value representing attitude; and/or wherein the attitude of an agent may be represented by a dynamic mathematical model that takes into account a plurality of factors influencing the attitude of a driver, the plurality of factors including one or any combination of age, type of vehicle, time of day, and traffic conditions.
8. A collision prediction apparatus according to one or more of the preceding claims, further comprising: a controller configured to specify a set of input parameters and to instruct the simulator to execute a number of iterations with the specified input parameters, wherein the input parameters include one or more of the following: the dimensions and/or layout of the road represented by the simulation environment; the visibility conditions in the simulation environment; the density of agents in the simulation environment and/or the average speed of the agents; the visibility of each agent in the simulation environment; a proportion of further agents belonging to each of a predetermined set of categories of agent, including bicycle, car, HGV, and/or motorcycle; a category of agent to which the subject agent belongs from among a predetermined set of categories of agent, including bicycle, car, HGV, and/or motorcycle; if the subject agent belongs to the category of bicycle, other characteristics including gender of rider, and whether or not the rider is wearing a helmet, an attitude characteristic of the agent, and/or distance between the agent and the kerb; wherein the set of piles attributed to each agent in the initiation phase is configurable according to the input parameters.
9. A collision prediction apparatus according to claim 8, wherein the apparatus further comprises a road layout selection module, being configured to generate the (r)simulation environment in a plurality of versions, each version differing from one another in dimensions and/or r layout, and to output the generated simulation environments to the simulator; the simulator being configured to execute a predetermined number of iterations for each version of the CD simulation environment, with any other input parameters being the same for each version; the road layout selection module being further configured to receive the value output by the simulator for each version of the simulation environment, and to identify and output the version of the simulation environment having the lowest collision rate as a suggested road layout.
10. A collision prediction apparatus according to any of the preceding claims, further comprising: a graphical representation generator configured to generate and output a graphical representation of the simulation environment and the agents in the simulation environment.
11. A collision prediction apparatus according to claim 1, wherein the collision prediction apparatus is a component of a vehicle; the collision prediction apparatus further comprises a real-time data receiving module configured to receive real-time data representing the density of vehicles on a particular section of road and/or the average speed of the vehicles from a monitoring apparatus as the vehicle approaches the particular section of road; and a road layout characteristic receiving module configured to receive road layout data representing the dimensions and layout of the particular section of road from the monitoring apparatus as the vehicle approaches the particular section of road; the controller is configured to input the real-time data to the simulator as input parameters, the simulator being configured to execute a simulation in which the density of agents and/or average agent speeds in the simulation correspond to the real-time data, and the simulation environment is configured to represent the particular section of road according to the received road layout data; and the alert signal causes a warning to be generated within the vehicle.
12. A computer-implemented collision prediction method, the method comprising, at a computing apparatus: executing an iterative agent-based simulation in which, in each iteration: a subject agent representing a travelling party travels from an end of a simulation environment representing a thoroughfare toward another end; one or more further agents representing vehicles interact with the subject agent, the iterative agent-based simulation being configured to categorise the outcome of each interaction as either a collision or not a collision; r the iteration being complete when the subject agent either reaches the another end, or is involved in an interaction of which the outcome is categorised as a collision; CD the method further comprising: recording and outputting a collision rate value representing a number of interactions in which the outcome is categorised as a collision per iteration; and generating and outputting an alert signal when the collision rate value output by the simulator satisfies an alert criterion, wherein: the simulation environment is based on a particular section of road, and real-time data representing the density of vehicles on the particular section of road and/or the average speed of the vehicles is monitored by a monitoring module, and the method includes inputting the real-time data to the iterative agent-based simulation as input parameters, the executed simulation including density of agents and/or average agent speeds corresponding to the real-time data, and the simulation environment represents the particular section of road; and the alert signal causes configurable signage located at the particular section of road to display a warning.
13. A computer program which, when executed by a computing apparatus, causes the computing apparatus to function as the collision prediction apparatus as claimed in any of claims 1 to 11.