CN111489585A - Vehicle and pedestrian collision avoidance method based on edge calculation - Google Patents

Abstract

The invention discloses a vehicle and pedestrian collision avoidance method based on edge calculation, which belongs to the field of Internet of things, and is characterized in that under the environment of edge calculation, communication between a vehicle and pedestrian equipment is realized through an edge server, communication delay is reduced, the position of the vehicle is predicted by Kalman filtering, the pedestrian track is predicted by using a sensor on the pedestrian equipment, the limitation of visual range is overcome, the collision detection precision is improved, the method is suitable for most non-visual range (N L OS) scenes, a collision risk model based on collision time (TTC) is established according to the relative motion relationship between the vehicle and the pedestrian, and simulation results show that the collision avoidance algorithm provided by the invention can effectively reduce the collision risk of the vehicle and the pedestrian.

Description

Vehicle and pedestrian collision avoidance method based on edge calculation

Technical Field

The invention relates to the technical field of Internet of things, in particular to a vehicle and pedestrian collision avoidance method based on edge calculation.

Background

In 2017, Hassan Artail proposes a scheme for realizing pedestrian and vehicle communication by using a cellular technology in Avoding car-pedestrian communication using a VANETto cellular communication frame, so as to detect possible vehicle and pedestrian collision. In the technical scheme, the remote cloud server is used as the intermediate server to indirectly realize the communication between the vehicle and the pedestrian, so that the limitation of the sight distance is overcome, but the system delay is very high and is difficult to determine, which is undoubtedly fatal to a pedestrian safety algorithm program extremely sensitive to delay; moreover, the pedestrian and vehicle position prediction algorithm in the technical scheme is extremely rough, all positions where pedestrians and vehicles possibly appear simultaneously can be marked as unsafe regions under most conditions, and the false alarm probability is high.

Disclosure of Invention

The invention aims to provide a vehicle and pedestrian collision avoidance method based on edge calculation, which not only reduces communication delay between vehicles and pedestrians, but also improves the early warning accuracy rate of vehicle and pedestrian collision accidents.

The technical purpose of the invention is realized by the following technical scheme:

a vehicle pedestrian collision avoidance method based on edge calculation, comprising the steps of:

step 1: establishing a network model of pedestrian device-to-vehicle communication: arranging a Road Side Unit (RSU)/base station and a Mobile Edge Computing (MEC) server connected with the RSU/base station at an intersection or a road edge, wherein the pedestrian equipment is a smart phone with a GPS module, and the pedestrian equipment and the vehicle can realize communication in the coverage range of the network model;

in the mode, the MEC server can be connected to a roadside unit (RSU)/base station, a vehicle can communicate with the RSU/base station through L TE-V2X, and a pedestrian can communicate with the RSU/base station through a smart phone;

step 2: respectively predicting the future positions of the vehicles and the pedestrians by utilizing a Kalman filter according to the current information and the historical information of the vehicle and pedestrian devices;

step 3: and according to the predicted paths of the vehicles and the pedestrians in Step2, running a vehicle and pedestrian collision avoidance algorithm according to the mutual motion relation, and early warning the possible collision.

Further, the specific steps of predicting the future position of the vehicle in Step2 are as follows:

assuming that the automobile has constant rotation rate and speed, the state vector of the automobile is selected as follows:

X＝[x y v θ ω]^T

theta is a yaw angle which is an included angle between a tracked target vehicle and an x axis under a current vehicle coordinate system, the anticlockwise direction is positive, the numeric area is [0, 2 pi ], and omega is a yaw angular speed; the moving state of the running automobile is generally straight or turning due to the limitation of roads, traffic regulations and the like, the two moving states can be represented by the state vector, and when omega is not equal to 0, the automobile runs in a non-straight line, and the state transfer function of the model is as follows:

when ω is 0, the model is driven in a straight line, and the state transition function of the model becomes:

for the multivariate functions of (1) and (2), a multivariate Taylor series is used for expansion, high-order series is omitted, only the linearization by a Jacobian matrix is considered, and in the model, partial derivatives of each element can be obtained to obtain a Jacobian matrix J_AWhen ω ≠ 0:

when ω is 0, the jacobian matrix is:

calculating uncertainty Q caused by noise processing; in the network model, the introduction of noise mainly originates from two places: linear acceleration and yaw rate noise that affect the state quantity (x, y, v, θ, ω), the linear acceleration and yaw rate noise having the following expression for the effect on the state:

wherein mu_a，

For acceleration on a straight line and on a corner, Q is a covariance matrix for processing noise, and is expressed as:

Q＝E[n·n^T]＝E[Gμμ^TG^T]＝G·E[μμ^T]·G^T(4)

wherein

Therefore, the calculation formula of the covariance matrix Q for processing noise in the CTRV model is:

selecting data collected by two sensors, namely a laser radar and a radar, to perform sensor fusion to predict the vehicle track when a measurement equation is established;

firstly, the laser radar, the data of which is in a Cartesian coordinate system, can detect the position of a vehicle, but has no speed information, and the measured value of the laser radar is the coordinate (x, y) of a target vehicle; therefore, the measurement model of the lidar is still linear, and the measurement matrix is:

the corresponding measurement equation is:

secondly, radar, the measured target vehicle data of which are under a polar coordinate system, the prediction mapping of the radar to the measurement space is nonlinear^[16]The expression is as follows:

when h (x) is used to represent such a non-linear mapping, then the non-linear process is also linearized using taylor's formula when solving for kalman gain, and the jacobian matrix of h (x) is solved with reference to the prediction process:

predicting and correcting unknown parameters J in a model_H，J_AAfter Q and R (processing noise, generally provided by the manufacturer) have been solved, the data can be filtered according to the network model, so as to predict the future position of the vehicle.

Further, the specific steps of determining the future position of the pedestrian in Step2 are as follows:

the position of the pedestrian is determined according to the information collected by the GPS module of the smart phone, but many errors exist in the dynamic positioning of the GPS module, such as satellite clock error and the like; in order to eliminate errors and improve positioning accuracy, a Kalman filtering method is applied to a current statistical model of GPS dynamic filtering;

starting with a positioning result output by a GPS receiver directly, the total error caused by various error sources of GPS positioning in all directions is equivalent to the sum of a current mean value and colored noise conforming to a first-order Markov process; according to the 'current' statistical model, the state vector of the pedestrian is selected as:

the state equation for the system may be defined as:

wherein, A is a system state transition matrix:

A＝diag[A_x，A_y，A_z]

A_y，A_zand so on;

w (t) is:

＝[0 0 W_axW_x0 0 W_ayW_y0 0 W_azW_z]^T

is a system noise vector;

τ_ax，τ_ay，τ_azrespectively, acceleration-dependent time constants; tau is_x，τ_y，τ_zRespectively, the relevant time constants of the corresponding Markov processes; w_ax，W_ay，W_azAre respectively as

White gaussian noise of (1);

is the average value of the current acceleration on three coordinate axes;

taking the GPS receiver in 3 coordinate axis directions when establishing a measurement equationUpward positioning result L_x，L_y，L_zIs a measured value, then:

L_x＝x+∈_x+ω_lx

L_y＝y+∈_y+ω_ly

L_z＝z+∈_z+ω_lz

writing in matrix form:

L＝[L_x，L_y，L_z]^T

v is observation noise V ═ omega_lx，ω_ly，ω_lz]^T

H is an observation equation

The corresponding measurement equation is

L＝HX+V (7)

So far, we use the observation equation of GPS system to establish the measurement equation of system, use the "current" statistical model and the error model of GPS system to establish the state transition equation of system, and then according to the above system equation and observation equation, set the sampling period as T, through the discretization process^[18]The discrete kalman filter equation is established as follows:

K(k+1)＝P(k+1，k)H^T(k+1)[H(k+1)P(k+1)H^T(k+1)+R(k+1)]^-1

(9)

P(k+1，k)＝Φ(k+1，k)P(k)Φ^T(k+1，k)+Q(k) (10)

P(k+1)＝[I-K(k+1)H(k+1)]P(k+1，k) (11)

wherein phi_l(k+1，k)＝

diag[Φ_lx(k+1，k)，Φ_ly(k+1，k)，Φ_lz(k+1，k)]

Φ_ly，Φ_lzAnd so on;

Φ (k +1, k) in equation (10) is the discretization matrix of the system transfer matrix a:

Φ(k+1，k)＝diag[Φ_x(k+1，k)，Φ_y(k+1，k)，Φ_z(k+1，k)]

Φ_y(k+1，k)，Φ_z(k +1, k) and so on, simply let τ in (12)_ax，τ_xRespectively converted into tau_ay，τ_yAnd τ_az，τ_zThen the method is finished;

q (k) is a discretization matrix of a systematic noise covariance matrix Q

R (k) is a discretization matrix of an observation noise covariance matrix R of the system

Q (k), R (k) are positive definite matrixes;

and (4) until the unknown parameters in the prediction and correction models are solved, and then performing corresponding data processing according to a conventional Kalman filtering program.

Further, the Step3 is a vehicle and pedestrian collision avoidance algorithm, which comprises the following specific steps:

step3.1: the vehicle and pedestrian equipment periodically collect respective state information through a sensor, upload the state information to an MEC server through an RSU/base station, and operate a corresponding Kalman filter on the MEC server according to the state information of pedestrians and vehicles to predict the tracks of traveling people and vehicles;

in Step3.1, the motion of the pedestrian needs to be regarded as nonlinear because the motion of the pedestrian has great uncertainty, and EKF is a filter for predicting the position of a nonlinear moving object; compared with pedestrians, vehicles running on the road can roughly be divided into two classes due to the limitation of the road, wherein the two classes are straight running, the yaw rate is zero, and the vehicles can be regarded as linear motion; one is that the yaw rate is not zero at the time of turning and the motion is nonlinear;

step3.2: running a collision avoidance algorithm to calculate Time To Collision (TTC) and determining a likelihood of a vehicle pedestrian collision from the value of TTC to alert the vehicle driver and the pedestrian;

the TTC (time to collision) is as follows:

L₁，L₂the lengths of the trajectories to the collision points, V, for vehicles and pedestrians, respectively₁，V₂Current speeds of the vehicle and pedestrian, t_v，t_pRespectively, communication delay from the vehicle and the pedestrian to the edge server, and judging whether collision occurs according to the TTC value;

when TTC is 0, that is, if the current vehicle and pedestrian keep running without taking measures, a collision may occur, at which time, a warning message is transmitted to the vehicle, and the vehicle is notified of taking measures before an accident occurs, thereby avoiding the collision.

Further, in step3.2, since there is an unavoidable error in the algorithm, ttcr (time to collision range) and the threshold ξ are set when

TTCR≤ξ

Collision early warning is carried out, so that the influence of errors and vehicle length generated by track prediction is eliminated.

The invention has the beneficial effects that:

firstly, the feasible communication mode of the vehicles and the pedestrians under the current network environment is established through the MEC server, the communication delay is effectively reduced, the delay from the vehicles to the MEC server is less than 10ms, and the delay from the pedestrian equipment to the server is less than 50ms, so that the accuracy of the vehicle and pedestrian collision accident early warning is improved, and the probability of the vehicle and pedestrian collision accidents under the non-line-of-sight conditions such as rain and fog road blind areas is effectively reduced.

Secondly, the feasible communication mode of vehicles and pedestrians in the current network environment is established through the MEC server, the burden of the cloud server is greatly relieved by introducing the MEC, the efficiency of the system is improved, and the throughput of the system is increased.

Thirdly, the invention introduces the pedestrian device, namely the smart phone, overcomes the non-line-of-sight limitation, effectively expands the application scene, adds the device which is carried by the pedestrian and can provide the location service at any time and any place into the Internet of vehicles, obviously improves the safety of the non-motor road users, and the introduction of the smart phone has a positive influence on the current situation that the current pedestrian safety research takes the vehicle as the center of gravity.

Fourthly, the invention also adopts Kalman filtering to predict the future positions of the vehicles and the pedestrians, thereby further improving the accuracy of the vehicle and pedestrian collision accident early warning.

Drawings

FIG. 1 is a diagram of a system model of the present invention;

FIG. 2 is a graph of data results showing the effect of vehicle speed on collision warning success;

FIG. 3 is a graph of data results showing the effect of pedestrian velocity on collision warning success;

FIG. 4 is a graph of comparative experimental data results.

Detailed Description

For the purpose of describing the invention in more detail and facilitating understanding for those skilled in the art, the present invention will be further described with reference to the accompanying drawings and examples, which are provided for illustration and understanding of the present invention and are not intended to limit the present invention.

Example (b):

a vehicle pedestrian collision avoidance method based on edge calculation, as shown in fig. 1, comprising the steps of:

step 1: establishing a network model of pedestrian device-to-vehicle communication: arranging a Road Side Unit (RSU)/base station and a Mobile Edge Computing (MEC) server connected with the RSU/base station at an intersection or a road edge, wherein the pedestrian equipment is a smart phone with a GPS module, and the pedestrian equipment and the vehicle can realize communication in the coverage range of the network model;

in the mode, the MEC server can be connected to a roadside unit (RSU)/base station, a vehicle can communicate with the RSU/base station through L TE-V2X, and a pedestrian can communicate with the RSU/base station through a smart phone;

step 2: respectively predicting the future positions of the vehicles and the pedestrians by utilizing a Kalman filter according to the current information and the historical information of the vehicle and pedestrian devices;

step 3: and according to the predicted paths of the vehicles and the pedestrians in Step2, running a vehicle and pedestrian collision avoidance algorithm according to the mutual motion relation, and early warning the possible collision.

Further, the specific steps of predicting the future position of the vehicle in Step2 are as follows:

assuming that the automobile has constant rotation rate and speed, the state vector of the automobile is selected as follows:

X＝[x y v θ ω]^T

theta is a yaw angle which is an included angle between a tracked target vehicle and an x axis under a current vehicle coordinate system, the anticlockwise direction is positive, the numeric area is [0, 2 pi ], and omega is a yaw angular speed; the moving state of the running automobile is generally straight or turning due to the limitation of roads, traffic regulations and the like, the two moving states can be represented by the state vector, and when omega is not equal to 0, the automobile runs in a non-straight line, and the state transfer function of the model is as follows:

when ω is 0, the model is driven in a straight line, and the state transition function of the model becomes:

for the multivariate functions of (1) and (2), a multivariate Taylor series is used for expansion, high-order series is omitted, only the linearization by a Jacobian matrix is considered, and in the model, partial derivatives of each element can be obtained to obtain a Jacobian matrix J_AWhen ω ≠ 0:

when ω is 0, the jacobian matrix is:

calculating uncertainty Q caused by noise processing; in the network model, the introduction of noise mainly originates from two places: linear acceleration and yaw rate noise that affect the state quantity (x, y, v, θ, ω), the linear acceleration and yaw rate noise having the following expression for the effect on the state:

wherein mu_a，

For acceleration in a straight line and in a corner, Q is a co-ordinate for noise handlingA variance matrix, whose expression is:

Q＝E[n·n^T]＝E[Gμμ^TG^T]＝G·E[μμ^T]·G^T(4)

wherein

Therefore, the calculation formula of the covariance matrix Q for processing noise in the CTRV model is:

selecting data collected by two sensors, namely a laser radar and a radar, to perform sensor fusion to predict the vehicle track when a measurement equation is established;

firstly, the laser radar, the data of which is in a Cartesian coordinate system, can detect the position of a vehicle, but has no speed information, and the measured value of the laser radar is the coordinate (x, y) of a target vehicle; therefore, the measurement model of the lidar is still linear, and the measurement matrix is:

the corresponding measurement equation is:

secondly, radar, the measured target vehicle data of which are under a polar coordinate system, the prediction mapping of the radar to the measurement space is nonlinear^[16]The expression is as follows:

when h (x) is used to represent such a non-linear mapping, then the non-linear process is also linearized using taylor's formula when solving for kalman gain, and the jacobian matrix of h (x) is solved with reference to the prediction process:

predicting and correcting unknown parameters J in a model_H，J_AAfter Q and R (processing noise, generally provided by the manufacturer) have been solved, the data can be filtered according to the network model, so as to predict the future position of the vehicle.

Further, the specific steps of determining the future position of the pedestrian in Step2 are as follows:

the position of the pedestrian is determined according to the information collected by the GPS module of the smart phone, but many errors exist in the dynamic positioning of the GPS module, such as satellite clock error and the like; in order to eliminate errors and improve positioning accuracy, a Kalman filtering method is applied to a current statistical model of GPS dynamic filtering;

starting with a positioning result output by a GPS receiver directly, the total error caused by various error sources of GPS positioning in all directions is equivalent to the sum of a current mean value and colored noise conforming to a first-order Markov process; according to the 'current' statistical model, the state vector of the pedestrian is selected as:

the state equation for the system may be defined as:

wherein, A is a system state transition matrix:

A＝diag[A_x，A_y，A_z]

A_y，A_zand so on;

w (t) is:

＝[0 0 W_axW_x0 0 W_ayW_y0 0 W_azW_z]^T

is a system noise vector;

τ_ax，τ_ay，τ_azrespectively, acceleration-dependent time constants; tau is_x，τ_y，τ_zRespectively, the relevant time constants of the corresponding Markov processes; w_ax，W_ay，W_azAre respectively as

White gaussian noise of (1);

is the average value of the current acceleration on three coordinate axes;

the positioning result L of the GPS receiver in 3 coordinate axis directions is taken when the measurement equation is established_x，L_y，L_zIs a measured value, then:

L_x＝x+∈_x+ω_lx

L_y＝y+∈_y+ω_ly

L_z＝z+∈_z+ω_lz

writing in matrix form:

L＝[L_x，L_y，L_z]^T

v is observation noise V ═ omega_lx，ω_ly，ω_lz]^T

H is an observation equation

The corresponding measurement equation is L HX + V (7)

So far, we use the observation equation of GPS system to establish the measurement equation of system, use the "current" statistical model and the error model of GPS system to establish the state transition equation of system, and then according to the above system equation and observation equation, set the sampling period as T, through the discretization process^[18]The discrete kalman filter equation is established as follows:

K(k+1)＝P(k+1，k)H^T(k+1)[H(k+1)P(k+1)H^T(k+1)+R(k+1)]^-1

(9)

P(k+1，k)＝Φ(k+1，k)P(k)Φ^T(k+1，k)+Q(k) (10)

P(k+1)＝[I-K(k+1)H(k+1)]P(k+1，k) (11)

wherein phi_l(k+1，k)＝

diag[Φ_lx(k+1，k)，Φ_ly(k+1，k)，Φ_lz(k+1，k)]

Φ_ly，Φ_lzAnd so on;

Φ (k +1, k) in equation (10) is the discretization matrix of the system transfer matrix a:

Φ(k+1，k)＝diag[Φ_x(k+1，k)，Φ_y(k+1，k)，Φ_z(k+1，k)]

Φ_y(k+1，k)，Φ_z(k +1, k) and so on, it is only necessary to add to (12)τ_ax，τ_xRespectively converted into tau_ay，τ_yAnd τ_az，τ_zThen the method is finished;

q (k) is a discretization matrix of a systematic noise covariance matrix Q

R (k) is a discretization matrix of an observation noise covariance matrix R of the system

Q (k), R (k) are positive definite matrixes;

and (4) until the unknown parameters in the prediction and correction models are solved, and then performing corresponding data processing according to a conventional Kalman filtering program.

Further, the Step3 is a vehicle and pedestrian collision avoidance algorithm, which comprises the following specific steps:

step3.1: the vehicle and pedestrian equipment periodically collect respective state information through a sensor, upload the state information to an MEC server through an RSU/base station, and operate a corresponding Kalman filter on the MEC server according to the state information of pedestrians and vehicles to predict the tracks of traveling people and vehicles;

in Step3.1, the motion of the pedestrian needs to be regarded as nonlinear because the motion of the pedestrian has great uncertainty, and EKF is a filter for predicting the position of a nonlinear moving object; compared with pedestrians, vehicles running on the road can roughly be divided into two classes due to the limitation of the road, wherein the two classes are straight running, the yaw rate is zero, and the vehicles can be regarded as linear motion; one is that the yaw rate is not zero at the time of turning and the motion is nonlinear;

step3.2: running a collision avoidance algorithm to calculate Time To Collision (TTC) and determining a likelihood of a vehicle pedestrian collision from the value of TTC to alert the vehicle driver and the pedestrian;

the TTC (time to collision) is as follows:

L₁，L₂the lengths of the trajectories to the collision points, V, for vehicles and pedestrians, respectively₁，V₂Current speeds of the vehicle and pedestrian, t_v，t_pRespectively, communication delay from the vehicle and the pedestrian to the edge server, and judging whether collision occurs according to the TTC value;

when TTC is 0, that is, if the current vehicle and pedestrian keep running without taking measures, a collision may occur, at which time, a warning message is transmitted to the vehicle, and the vehicle is notified of taking measures before an accident occurs, thereby avoiding the collision.

Further, in step3.2, since there is an unavoidable error in the algorithm, ttcr (time to collision range) and the threshold ξ are set when

TTCR≤ξ

Collision early warning is carried out, so that the influence of errors and vehicle length generated by track prediction is eliminated.

As can be seen from fig. 2, the higher the vehicle speed is at TTC 0, the lower the warning accuracy rate, because the higher the vehicle speed is, the greater the error and delay have an effect thereon, when the threshold ξ is increased, the warning accuracy rate is first increased and reaches the highest value because the influence of the system error is gradually reduced after the threshold is increased.

It can also be seen that the warning accuracy curve is steeper as the vehicle speed is higher, and smoother as the vehicle speed is lower. Because the higher the vehicle speed, the more sensitive it is to delay, the lower the vehicle speed, the more accurate the trajectory prediction, and the corresponding accuracy is higher.

It can be seen from fig. 3 that the influence of the pedestrian speed and the vehicle speed on the system is approximately the same, and the curve trend also follows the trend of "steeper" speed and "gentler" speed, when the TTC is 0, the speed and the early warning success rate are inversely related, except that when the threshold ξ is the same, the early warning accuracy rate of the part is generally higher than that in 5.3, because the influence on the system is not as obvious as that of the vehicle because the pedestrian speed is slower than that of the vehicle.

FIG. 4 is a graph of the results of a comparative test in which the collision points are calculated based on the pedestrian trajectories of vehicles, and each pedestrian and vehicle has their own trajectory, so that the total number of vehicles and pedestrians will affect the performance of the system. this section studies the impact of the total number of pedestrians and vehicles on the system, and compares this with the method in the Avoding car-traffic utilization a VANETto cellular communication framework.

And a single vehicle or a pedestrian may receive a plurality of early warnings when the track increases, the total number of the early warnings is increased at this time, and after a certain vehicle receives accurate early warning information and decelerates successfully to avoid collision, other received early warning information is invalidated, namely, the number of early warning success times of the single vehicle or the pedestrian is less than or equal to 1, namely, the upper limit of the number of early warning success times is smaller, but only the number of early warning success times is larger and larger along with the increase of the total number and the upper limit is very high, so that the change trend of the early warning success rate is gradually reduced along with the increase of the total number of. Compared with the algorithm in the Avoiding car-companion simulation using a VANET cellular communication frame, the early warning success rate of the algorithm is higher.

While the present invention has been described in detail with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, and various changes can be made without departing from the spirit and scope of the present invention.

Claims (5)

1. A vehicle pedestrian collision avoidance method based on edge calculation is characterized in that: the method comprises the following steps:

step 1: establishing a network model of pedestrian device-to-vehicle communication: arranging a Road Side Unit (RSU)/base station and a Mobile Edge Computing (MEC) server connected with the RSU/base station at an intersection or a road edge, wherein the pedestrian equipment is a smart phone with a GPS module, and the pedestrian equipment and the vehicle can realize communication in the coverage range of the network model;

step 2: respectively predicting the future positions of the vehicles and the pedestrians by utilizing a Kalman filter according to the current information and the historical information of the vehicle and pedestrian devices;

step 3: and according to the predicted paths of the vehicles and the pedestrians in Step2, running a vehicle and pedestrian collision avoidance algorithm according to the mutual motion relation, and early warning the possible collision.

2. The vehicle pedestrian collision avoidance method based on the edge calculation according to claim 1, characterized in that: the concrete steps of predicting the future position of the vehicle in Step2 are as follows:

assuming that the automobile has constant rotation rate and speed, the state vector of the automobile is selected as follows:

X＝[x y v θ ω]^T

theta is a yaw angle which is an included angle between a tracked target vehicle and an x axis under a current vehicle coordinate system, the anticlockwise direction is positive, the numeric area is [0, 2 pi ], and omega is a yaw angular speed; the moving state of the running automobile is generally straight or turning due to the limitation of roads, traffic regulations and the like, the two moving states can be represented by the state vector, and when omega is not equal to 0, the automobile runs in a non-straight line, and the state transfer function of the model is as follows:

when ω is 0, the model is driven in a straight line, and the state transition function of the model becomes:

for the multivariate functions of (1) and (2), a multivariate Taylor series is used for expansion, high-order series is omitted, only the linearization by a Jacobian matrix is considered, and in the model, partial derivatives of each element can be obtained to obtain a Jacobian matrix J_AWhen ω ≠ 0:

when ω is 0, the jacobian matrix is:

calculating uncertainty Q caused by noise processing; in the network model, the introduction of noise mainly originates from two places: linear acceleration and yaw rate noise that affect the state quantity (x, y, v, θ, ω), the linear acceleration and yaw rate noise having the following expression for the effect on the state:

wherein mu_a，

For acceleration on a straight line and on a corner, Q is a covariance matrix for processing noise, and is expressed as:

Q＝E[n·n^T]＝E[Gμμ^TG^T]＝G·E[μμ^T]·G^T(4)

wherein

Therefore, the calculation formula of the covariance matrix Q for processing noise in the CTRV model is:

selecting data collected by two sensors, namely a laser radar and a radar, to perform sensor fusion to predict the vehicle track when a measurement equation is established;

firstly, the laser radar, the data of which is in a Cartesian coordinate system, can detect the position of a vehicle, but has no speed information, and the measured value of the laser radar is the coordinate (x, y) of a target vehicle; therefore, the measurement model of the lidar is still linear, and the measurement matrix is:

the corresponding measurement equation is:

secondly, radar, the measured target vehicle data of which are under a polar coordinate system, the prediction mapping of the radar to the measurement space is nonlinear^[16]The expression is as follows:

when h (x) is used to represent such a non-linear mapping, then the non-linear process is also linearized using taylor's formula when solving for kalman gain, and the jacobian matrix of h (x) is solved with reference to the prediction process:

predicting and correcting unknown parameters J in a model_H，J_AAfter Q and R (processing noise, generally provided by the manufacturer) have been solved, the data can be filtered according to the network model, so as to predict the future position of the vehicle.

3. The vehicle pedestrian collision avoidance method based on the edge calculation according to claim 1, characterized in that: the specific steps of determining the future position of the pedestrian in Step2 are as follows:

the position of the pedestrian is determined according to the information collected by the GPS module of the smart phone, but many errors exist in the dynamic positioning of the GPS module, such as satellite clock error and the like; in order to eliminate errors and improve positioning accuracy, a Kalman filtering method is applied to a current statistical model of GPS dynamic filtering;

starting with a positioning result output by a GPS receiver directly, the total error caused by various error sources of GPS positioning in all directions is equivalent to the sum of a current mean value and colored noise conforming to a first-order Markov process; according to the 'current' statistical model, the state vector of the pedestrian is selected as:

the state equation for the system may be defined as:

wherein, A is a system state transition matrix:

A＝diag[A_x，A_y，A_z]

A_y，A_zand so on;

w (t) is:

＝[0 0 W_axW_x0 0 W_ayW_y0 0 W_azW_z]^T

is a system noise vector;

τ_ax，τ_ay，τ_azrespectively, acceleration-dependent time constants; tau is_x，τ_y，τ_zRespectively, the relevant time constants of the corresponding Markov processes; w_ax，W_ay，W_azAre respectively as

White gaussian noise of (1);

is the average value of the current acceleration on three coordinate axes;

the positioning result L of the GPS receiver in 3 coordinate axis directions is taken when the measurement equation is established_x，L_y，L_zIs a measured value, then:

L_x＝x+∈_x+ω_lx

L_y＝y+∈_y+ω_ly

L_z＝z+∈_z+ω_lz

writing in matrix form:

L＝[L_x，L_y，L_z]^T

v is observation noise V ═ omega_lx，ω_ly，ω_lz]^T

H is an observation equation

The corresponding measurement equation is

L＝HX+V (7)

So far, we use the observation equation of GPS system to establish the measurement equation of system, use the "current" statistical model and the error model of GPS system to establish the state transition equation of system, and then according to the above system equation and observation equation, set the sampling period as T, through the discretization process^[18]The discrete kalman filter equation is established as follows:

K(k+1)＝P(k+1，k)H^T(k+1)[H(k+1)P(k+1)H^T(k+1)+R(k+1)]^-1

(9)

P(k+1，k)＝Φ(k+1，k)P(k)Φ^T(k+1，k)+Q(k) (10)

P(k+1)＝[I-K(k+1)H(k+1)]P(k+1，k) (11)

wherein phi_l(k+1，k)＝

diag[Φ_lx(k+1，k)，Φ_ly(k+1，k)，Φ_lz(k+1，k)]

Φ_ly，Φ_lzAnd so on;

Φ (k +1, k) in equation (10) is the discretization matrix of the system transfer matrix a:

Φ(k+1，k)＝diag[Φ_x(k+1，k)，Φ_y(k+1，k)，Φ_z(k+1，k)]

Φ_y(k+1，k)，Φ_z(k +1, k) and so on, simply let τ in (12)_ax，τ_xRespectively converted into tau_ay，τ_yAnd τ_az，τ_zThen the method is finished;

q (k) is a discretization matrix of a systematic noise covariance matrix Q

R (k) is a discretization matrix of an observation noise covariance matrix R of the system

Q (k), R (k) are positive definite matrixes;

and (4) until the unknown parameters in the prediction and correction models are solved, and then performing corresponding data processing according to a conventional Kalman filtering program.

4. The vehicle pedestrian collision avoidance method based on the edge calculation according to claim 1, characterized in that: the method for avoiding the collision of the vehicles and the pedestrians in Step3 comprises the following specific steps:

step3.1: the vehicle and pedestrian equipment periodically collect respective state information through a sensor, upload the state information to an MEC server through an RSU/base station, and operate a corresponding Kalman filter on the MEC server according to the state information of pedestrians and vehicles to predict the tracks of traveling people and vehicles;

in Step3.1, the motion of the pedestrian needs to be regarded as nonlinear because the motion of the pedestrian has great uncertainty, and EKF is a filter for predicting the position of a nonlinear moving object; compared with pedestrians, vehicles running on the road can roughly be divided into two classes due to the limitation of the road, wherein the two classes are straight running, the yaw rate is zero, and the vehicles can be regarded as linear motion; one is that the yaw rate is not zero at the time of turning and the motion is nonlinear;

step3.2: running a collision avoidance algorithm to calculate Time To Collision (TTC) and determining a likelihood of a vehicle pedestrian collision from the value of TTC to alert the vehicle driver and the pedestrian;

the TTC (time to collision) is as follows:

L₁，L₂the lengths of the trajectories to the collision points, V, for vehicles and pedestrians, respectively₁，V₂Current speeds of the vehicle and pedestrian, t_v，t_pRespectively, communication delay from the vehicle and the pedestrian to the edge server, and judging whether collision occurs according to the TTC value;

when TTC is 0, that is, if the current vehicle and pedestrian keep running without taking measures, a collision may occur, at which time, a warning message is transmitted to the vehicle, and the vehicle is notified of taking measures before an accident occurs, thereby avoiding the collision.

5. The method as claimed in claim 4, wherein the TTCR (time to collision range) and the threshold ξ are set when the algorithm has unavoidable error in Step3.2

TTCR≤ξ

Collision early warning is carried out, so that the influence of errors and vehicle length generated by track prediction is eliminated.

Images