ES2636832B1 - VIDEO SURVEILLANCE SYSTEM BASED ON ANALYSIS OF SEQUENCES OF PICTURES GENERATED BY EVENTS - Google Patents

Abstract

Video surveillance system based on analysis of sequences of images generated by events # The invention relates to a video surveillance system based on analysis of sequences generated by events, comprising: one or more sensors (1) configured to receive information from an environment under surveillance; an event activation module (2) connected to the sensors (1), configured for reading, analyzing and processing the information received by the sensors (1), as well as for the definition of events associated with one or more threshold values of said information received by the sensors (1); one or more video cameras (3), connected to the event activation module (2) and configured by said module (2) for their control; and one or more video capturers (4), configured to record and store the information captured by the cameras (3) during one or more sequences associated with the events defined by the activation module (2).

Description

5

10

fifteen

twenty

25

30

35

DESCRIPTION

VIDEO SURVEILLANCE SYSTEM BASED ON ANALYSIS OF SEQUENCES OF PICTURES GENERATED BY EVENTS

FIELD OF THE INVENTION

The present invention is framed in the field of video surveillance, that is, of activity or presence detection technologies based on video analysis. More specifically, the invention relates to a video surveillance system configured to perform precision analysis of the captured images, where said analysis is applied to certain sequences of images or clips defined by different configurable events. The system of the invention also allows its combination with other traditional video detection and process systems, substantially improving their efficiency and accuracy.

BACKGROUND OF THE INVENTION

Motion detection systems by video analysis generally consist of video processing equipment that analyzes the changes that occur in a sequence of images, with the aim of detecting changes that represent moving elements. Examples of such moving elements may be people, vehicles, animals, etc.

Traditional motion detection systems analyze video continuously, which is a significant consumption of storage and information processing resources. In addition, this consumption increases with the resolution of the images since, at a higher resolution, it is necessary to process a larger number of pixels and, therefore, this requires a greater consumption of computing resources. Although different solutions have been proposed over time to mitigate this problem, the reality is that processing video in real time significantly limits, even today, the power of the analysis algorithms that can be applied or, alternatively, multiplies the computing resources necessary for image processing, thereby increasing the cost of the solution applied for that purpose.

5

10

fifteen

twenty

25

30

35

On the other hand, known video processing systems are generally difficult to configure. In them it is necessary to define different parameters such as sensitivity, which are not always intuitive and whose value must sometimes vary depending on the inevitable changes of the environment (for example, a landscape can change a lot between winter and spring).

Within the aforementioned known video surveillance systems, one usually distinguishes between two groups of dominant technologies:

- Motion detection through video processing:

Motion detection by video processing consists, essentially, in the comparison of successive images in a sequence or clip. The way to make the comparison is the one that differentiates the different products existing in the market. A basic comparison operation is the subtraction of two consecutive images, where a maximum difference or threshold is set and where, if said threshold is exceeded by a specific pixel, it is considered to have changed and that this is due to the movement of some element of the picture.

Subtraction of pixels, in practice, faces many problems. The fundamental one is that the value of a pixel is essentially a coding of the light reflected by a point, and this can change by phenomena as usual as the displacement of clouds. There are many alternatives to the "subtraction of pixels", many of them included in what are called "background subtraction" or "background modeling" techniques. These techniques consist of maintaining a background that is periodically updated, in order to adapt the system to changes in lighting, or others that occur in the scene monitored by the camera. To maintain the background model it is always necessary that the video processing is carried out continuously.

There are many methods that rely on background extraction systems to filter the false alarms of a system. A background extraction system is an advanced motion analysis, through which the scene is modeled dynamically. A generalized model of the scene is obtained that is used to detect the changes produced in the scene by moving objects (people, vehicles, doors, shadows, etc.). Traditionally, these methods produce many false alarms due to the changes

3

5

10

fifteen

twenty

25

30

35

that occur naturally on the scene (lighting, weather conditions, shadows, etc.). Many of them have designed a phase of filtering objects in which they use additional information from the scene that tries to reduce the number of false alarms.

The main problem with these types of methods is that they require an initialization and maintenance of the background model of the scene. When it comes to controlled scenarios, this type of analytics sufficiently solves the monitoring of the state of the cameras, but in real scenarios, they are insufficient. There are many studies that try to solve the problems derived from this type of analysis, but in all of them, the scenario being monitored ends up being limited in one way or another.

- Local and remote analysis:

Local and remote analysis solutions are usually divided into local analysis technologies (or "edge analysis") and remote (or centralized) analysis. Edge analysis occurs at the end, that is, as close as possible to the video source. The clearest example of edge analysis is the so-called smart cameras, which are capable of processing video. Therefore, for a camera to be intelligent it must have a processing unit or processor incorporated. For its part, the remote analysis implies that the processing is done outside the camera, in a device that must capture the flow or “stream” of video that it emits.

Another key factor in the different video surveillance technologies is the transmission of video since, if said transmission is carried out at a remote location, this implies a much higher resource consumption than transmitting it locally. This cost increases with the resolution of the transmitted images. The higher the resolution, the greater the bandwidth required to transmit them. The increase in the resolution of video cameras has, therefore, an important impact on the cost of the analysis of said video. In all cases it implies a greater consumption of machine resources and, secondly, in the case of remote analysis, a greater consumption of bandwidth to collect video from the cameras.

As described in the previous paragraphs, it is necessary, in the present technical field, to develop technical solutions that allow obtaining effective video surveillance systems, which substantially reduce the occurrence of false detection positives, but without excessive consumption of resources from computing for the analysis of

4

5

10

fifteen

twenty

25

30

35

Video sequences captured by the camera.

The present invention is oriented to solve the aforementioned needs, by means of a novel video surveillance system based on analysis of sequences of images generated by events, which resolves the limitations of known techniques.

BRIEF DESCRIPTION OF THE INVENTION

A main object of the present invention is, therefore, to provide video analysis means that have high precision and high capacity to avoid false positives, while maintaining a reduced consumption of computing resources intended for this purpose.

Said object of the invention is achieved by means of a video surveillance system based on analysis of sequences of images generated by events, which advantageously comprises:

- one or more sensors configured to receive information from an environment under surveillance;

- an event activation module connected to the sensors, said module being configured with hardware and software means for reading, analyzing and processing the information received by the sensors, as well as for the definition of events associated with one or more values threshold of said information received by the sensors;

- one or more video cameras, connected to the event activation module and configured by said module for their control;

- one or more video capturers, configured by hardware and / or software means to record and store the information captured by the cameras during one or more sequences associated with the events defined by the activation module.

This achieves a system that allows a plurality of events of interest to be established from the data of its sensors, so that video recording and analysis is restricted to the sequences associated with said events, so that computing resources and storage are substantially optimized, being focused only on those sequences of interest for monitoring. Likewise, a highly effective system is also achieved in the identification and validation of possible false positives, whose frequency of occurrence is usually high in traditional video surveillance systems.

5

10

fifteen

twenty

25

30

35

In a preferred embodiment of the invention, the sensors comprise one or more time, temperature, humidity, smoke, fire, motion sensors, image analyzers, volumetric presence detectors, laser detectors, sensor cable detectors, sound detectors , magnetic detectors, inductive loop detectors, pressure detectors, and / or motion or presence sensors by means of WiFi-based radar, or a radio frequency communication standard.

In another preferred embodiment of the invention, video capturers are equipped in the cameras themselves. Alternatively, video capturers may be equipped in a device external to said cameras, for example in a digital video recording device ("DVR").

In another preferred embodiment of the invention, the system comprises a server for processing and analyzing the video data recorded and / or stored by the capturers.

In another preferred embodiment of the invention, the server is connected to a complementary video demand module, configured to analyze a video sequence corresponding to that detected by the cameras, but at a different resolution than that analyzed by said server.

In another preferred embodiment of the invention, the server is connected to the sensors, the event activation module, the cameras and / or the capturers via a communication network. More preferably, the communications network comprises a cloud network connected to the internet, cable connection, WiFi or a mobile communication network.

In another preferred embodiment of the invention, the event activation module is located remotely with respect to the sensors and / or the cameras, through the communications network.

In another preferred embodiment of the invention, the sensors and / or the cameras are geolocated.

In another preferred embodiment of the invention, the processing server is connected to an automatic or manual event verification device.

6

5

10

fifteen

twenty

25

In another preferred embodiment of the invention, the verification device comprises a graphic video playback interface, and / or information associated with the classification criteria or threshold applied by the event activation module to a human user.

In another preferred embodiment of the invention, the system comprises a learning subsystem configured with hardware or software means, to establish criteria for classifying sequences, based on results obtained in previous events.

DESCRIPTION OF THE FIGURES

Figure 1 shows, for illustrative and non-limiting purposes, a scheme of the main elements of the video surveillance system of the invention, according to a preferred embodiment thereof.

NUMERICAL REFERENCES USED IN THE FIGURES

In order to help a better understanding of the technical characteristics of the invention, the aforementioned Figures are accompanied by a series of numerical references where, for illustrative and non-limiting purposes, the following is represented:

 (one)

 Sensors

 (2)

 Event Activation Module

 (3)

 Video-cameras

 (4)

 Captors

 (5)

 Server

 (6)

 Communications network

 (7)

 Automatic or manual verification device

 (8)

 Learning subsystem

 (9)

 Human user

DETAILED DESCRIPTION OF THE INVENTION

A detailed description of the invention, referring to a preferred embodiment thereof shown in Figure 1, is set forth below. As shown in said

5

10

fifteen

twenty

25

30

35

figure, the proposed video surveillance system is essentially based on one or more sensors (1) configured to receive information from the monitored environment. In principle, said sensors (1) can be of any known type, such as time, temperature, humidity, smoke, fire, image sensors (both in the visible spectrum and in the infrared band, microwave, etc.) or of movement / presence. In a possible embodiment of the invention, it is possible to use as a sensor (1) a motion detection device, configured to detect video sequences in which pixels may have changed and where, accordingly, it is presumed that there has been a movement. Other possible examples of motion detection sensors (1) usable within the scope of the invention are volumetric detectors, laser, sensor cable detectors, sound detectors, magnetic detectors, inductive loop detectors, pressure detectors, etc. Likewise, it is also possible to use motion or presence sensors using a WiFi-based radar, or a radio frequency communication standard, such as Bluetooth.

The sensors (1) used in the system are preferably connected to an event activation module (2), configured with hardware and / or software means for reading, analyzing and processing the information received by the sensors (1). In this way, the event activation module (2) will have, at all times, the data related to temperature, humidity, presence / movement detection, etc. generated by the sensors (1).

The event activation module (2) is configured with hardware and / or software means to set one or more thresholds for detecting the data obtained, so that when said thresholds are reached or exceeded, an "event" of analysis. Examples of the thresholds used can be, for example, certain values of temperature, humidity, luminosity, pixel difference in motion detectors, presence detection in a volumetric sensor, or even a certain time frequency, in the case of using a stopwatch or clock as sensor (1). Thus, depending on the threshold values chosen in a particular embodiment of the invention, the activation module (2) will be responsible for determining what conditions start a relevant event, for later analysis by the video surveillance system of the invention.

Thanks to the event activation module (2), the system thus allows greater efficiency compared to traditional systems, being able to select those sequences that are

8

5

10

fifteen

twenty

25

30

35

classified as "interesting" (activated by reaching the threshold value or values), and discard those sequences considered as not relevant.

As an additional element of the invention, the video surveillance system also includes one or more video cameras (3), connected to the event activation module (2) and configured by said module (2) for control during the sequences of analyzes considered relevant. Although the operation of the video cameras (3) of the system is independent of the operation of the sensors (1), the latter, in combination with the event activation module, will be responsible for operating as a controller or “trigger” of said cameras, causing them to analyze video clips only at those moments that are of interest, according to the threshold parameters set for each sensor (1). Thus, the invention allows the cameras (3) to be activated when a certain temperature is reached, or with a certain temporal frequency, or when a certain movement or presence is signaled by the sensors (1). With this, the system of the invention makes it possible to significantly reduce the size of the video information to be analyzed, restricting only the sequences associated with the events of interest.

For the analysis and processing of the video clips registered by the cameras (3), the system of the invention comprises one or more video capturers (4), responsible for recording and storing the information captured by the cameras (3) during the sequences associated with the events of interest. Said capturers (4) may be equipped in the cameras themselves (3) (for example in the form of capture cards), or be part of an external device (for example, a digital video recording device, or " DVR ”designated by its acronym in English). Also, the analysis of the captured sequences can be applied locally (by edge analysis) or remotely, for example by means of a video processing and analysis server (5) connected to a communications network (6), for example a network in the cloud connected to the internet, whether via cable connection, WiFi, a mobile communication network, etc. In an optional embodiment of the system of the invention, the event activation module (2) can also be located remotely and connected to the sensors (1) and / or to the cameras (3) through the communication network (6 ).

Remote information management, both to determine the events with sequences of interest (generated from the threshold values of the sensor information (1)),

9

5

10

fifteen

twenty

25

30

35

As for analyzing these sequences after recording by the capturers (4), it has some additional advantages, such as being able to centralize in a single machine the classification of video sequences from several remote locations, using, for example, the same server (5 ) of processing. The system of the invention also allows sharing different criteria for generating events between different locations, if desired. Likewise, it is also possible to use geolocated sensors (1) or cameras (3), so that their locations are uniquely determined. This allows, for example, to discriminate false positives generated by the presence of rain, when these occur in a subgroup of cameras (3) close to each other.

In a further preferred embodiment of the invention, the processing server (5) may be connected to an automatic or manual verification device (7), so that, within the sequences of interest, those detections corresponding to false positives, or are considered as non-relevant detections. In cases of manual verification, the invention offers a more efficient verification solution than that of known systems, since the human user (8) is only shown the videos that have been previously filtered as relevant, for exceeding the set threshold by the event activation module (2). Human intervention can be performed, for example, through a graphic video playback interface, and / or display information associated with the classification criteria or threshold applied by the event activation module (2) to the user (8) that performs Manual verification Thus, the invention allows classifying possible anomalies, since some may be considered relevant, while others may not be and will be discarded.

In automatic verification, on the other hand, an analysis of the pixels is preferably applied in the images that form the sequence of the event, where one or more specific classifiers will be assigned to recognize the different types of scene, depending on the activities to be monitored . Each classifier determines whether the scene belongs to the set of types that it knows or not.

In both the automatic and manual verification embodiments, the system of the invention may comprise a learning subsystem (9) configured by hardware or software means, which allows to establish criteria for classifying the sequences, based on results obtained in previous events. This allows to progressively improve the accuracy of the system and feed it back with the information previously obtained.

10

5

10

fifteen

twenty

25

30

35

The learning subsystem (9) can be connected to the verification device (7), being remotely connected to the sensors (1), the event activation module (2), the video cameras (3), and / or video capturers (4).

Finally, the system of the invention can comprise a complementary video demand module connected to the server (5), which allows analyzing a video sequence corresponding to that initially detected by the cameras (3) but at a different resolution (preferably higher) that analyzed by said server (5). Nowadays it is common to find cameras (3) that offer two or more video streams with different resolutions. The system of the invention, having its sensors (1) and the processing server (5) decoupled, allows working with video sequences that represent the same, but in multiple resolutions. This may allow applying classifiers on high resolution sequences, therefore on more information. This is especially important for certain types of classifiers, which require a more detailed analysis.

An example of the implementation of the system of the invention is, for example, the filtering of false alarms in video surveillance for perimeter protection. For this, a local team of intelligent cameras (3) with VAT (“intelligent video analysis”) functionality is used, with the aim of detecting movement or presence in the monitored perimeter. In practice, this type of system has problems with situations such as sudden light changes, heavy rain, occlusions or camera problems (3) (flickering image). All these types of situations generate false alarms that are received at an alarm receiving center (or CRA). For example, in the case of a day of heavy rain in a certain area, for a local CRA it can involve hundreds of alarms activated per hour, being actually false positives. The present invention would allow a user (8) or CRA operator, through the use of humidity detectors as sensors (1), to check for anomalous sequences that exceed a certain humidity threshold, and label these false positives as “rain alarm ”, So that the system automatically learns and classifies them thanks to the learning subsystem (9). Thanks to this classification, the operator can choose to ignore future alarms or treat them with less priority than other types of alarms. The example of rain is very common, but so are objects in the foreground, such as a fly or a bird. Being very close to the camera (3), these objects appear as huge elements. The invention also allows us to rule out this type of false positives, without incurring a large consumption of processing resources.

Likewise, the invention also makes it possible to take advantage of the increasing number of low-cost capturers (4), recorders or DVRs that are available on the market. Currently, there are DVR devices with integrated analytics that, without having advanced video analysis technology, have costs 90% lower than the most accurate equipment available. The invention 5 thus allows to take advantage of said low-cost DVR equipment with integrated analytics, which even generating false alarms a priori, can be corrected and reused thanks to the sequence filtering and verification provided by the system. This avoids having to have a second device with greater precision in the installation, avoiding hardware cost overruns and applying advanced analysis algorithms on the server (5), which can be both in the CRA and offered as a cloud service.

Claims (15)

5 10 fifteen twenty 25 30 1. - Video surveillance system based on analysis of sequences of images generated by events, characterized in that it comprises: - one or more sensors (1) configured to receive information from an environment under surveillance; - an event activation module (2) connected to the sensors (1), said module (2) being configured with hardware and software means for reading, analyzing and processing the information received by the sensors (1), thus as for the definition of events associated with one or more threshold values of said information received by the sensors (1); - one or more video cameras (3), connected to the event activation module (2) and configured by said module (2) for their control; Y - one or more video capturers (4), configured by hardware and / or software means to record and store the information associated with the images captured by the cameras (3), during one or more sequences associated with the events defined by the activation module (2). 2. - System according to the preceding claim, wherein the sensors (1) comprise one or more time, temperature, humidity, smoke, fire, motion sensors, image analyzers, volumetric presence detectors, laser detectors, cable detectors sensor, sound detectors, magnetic detectors, inductive loop detectors, pressure detectors, and / or motion or presence sensors using WiFi-based radar, or a standard for radio frequency communication. 3. - System according to any of the preceding claims, wherein the video capturers (4) are equipped in the cameras themselves (3). 4. - System according to any of claims 1-2, wherein the video capturers (4) are equipped in a device external to the cameras (3). 5. - System according to the preceding claim, wherein the video capturers (4) comprise a digital video recording device (DVR). 5 10 fifteen twenty 25 30 35 6. - System according to any of the preceding claims, comprising a server (5) for processing and analyzing the video data recorded and / or stored by the capturers (4). 7. - System according to the preceding claim, wherein the server (5) is connected to a complementary video demand module, configured to analyze a video sequence corresponding to that detected by the cameras (3), but at a different resolution than the one analyzed by said server (5). 8. - System according to any of claims 6-7, wherein the server (5) is connected to the sensors (1), the event activation module (2), the cameras (3) and / or the capturers ( 4) through a communications network (6). 9. - System according to the preceding claim, wherein the communications network (6) comprises a cloud network connected to the internet, cable connection, WiFi or a mobile communication network. 10. - System according to any of claims 8-9, wherein the event activation module (2) is located remotely with respect to the sensors (1) and / or the cameras (3), through the communications network (6). 11. - System according to any of the preceding claims, wherein the sensors (1) and / or the cameras (3) are geolocated. 12. - System according to any of the preceding claims, wherein the processing server (5) is connected to an automatic or manual event verification device (7). 13. - System according to the preceding claim, wherein the verification device (7) comprises a graphic video playback interface, and / or information associated with the classification criteria or threshold applied by the event activation module (2) to a human user (8). 14. - System according to any of the preceding claims, comprising a learning subsystem (9) configured with hardware or software means, for 14 establish criteria for classification of sequences, based on results obtained in previous events. 15. System according to the preceding claim, wherein the learning subsystem (9) 5 is connected to the verification device (7), and is connected remotely to the sensors (1), to the event activation module (2) , to video cameras (3), and / or video capturers (4).