Road Pedestrian Detection System Based on Computer Vision

Abstract

The present invention relates to the technical field of highway pedestrian detection, and specifically to a road pedestrian detection system based on computer vision. It includes a historical data module, which is used to collect pedestrians on the highway events in historical data, and trace the event process forward according to the time point of the pedestrians on the highway events, and analyze the pedestrians on the highway inducement; a video acquisition module, which collects video images on the highway and within a preset range around the highway at each monitoring point; an image recognition module, which recognizes the video images, extracts the characteristic information of the hidden danger events according to the video images, and recognizes the hidden danger events associated with the pedestrians on the highway inducement according to the characteristic information; an event judgment module, when there is a hidden danger event, judges the degree of fit of the current hidden danger event with the pedestrians on the highway inducement and the expected time of occurrence according to the characteristic information.

Description

Technical Field

[0001] The present invention relates to the technical field of highway pedestrian detection, and in particular to a road pedestrian detection system based on computer vision. Background Art

[0002] Expressways are roads designed for high-speed driving of cars, and the design speed is usually between 80 and 120 kilometers per hour. Therefore, it is very dangerous to have pedestrians on expressways. When pedestrians appear on expressways, it will cause serious casualties and economic losses. This special situation involving human life is also the most noteworthy issue in expressway management. Therefore, detecting pedestrians on expressways and issuing early warnings are of great significance to avoiding major traffic accidents on expressways.

[0003] In order to prevent pedestrians from entering the highway, the common practice is to set up warning signs and fences on the highway to prevent pedestrians from entering. In fact, most people also know not to enter the highway at will. However, the reality is that most people will ignore this point under special circumstances. For example, if a traffic accident occurs on the highway, the people on the car get off to check the accident. In a hurry, pedestrians will enter the highway. Another example is in bad weather conditions, such as heavy snow and fog, which affect people's judgment, and pedestrians will also enter the highway.

[0004] The existing personnel detection usually only detects personnel, that is, detects the position of personnel. When a person appears on the road, it is determined that a person has entered the highway. However, in this case, the behavior of entering the highway has already occurred, and it is too late to issue an early warning. Summary of the invention

[0005] The technical problem solved by the present invention is to provide a road pedestrian detection system based on computer vision, which can identify possible incidents of people entering the highway through feature analysis, thereby providing early warning.

[0006] The basic solution provided by the present invention: a road pedestrian detection system based on computer vision, characterized in that: it includes a server, and the server includes a historical data module, a video acquisition module, an image recognition module, an event judgment module and an early warning module;

[0007] The historical data module is used to collect the pedestrians on the highway events in the historical data, and trace the event process forward according to the time point of the pedestrians on the highway events, and analyze the causes of pedestrians on the highway;

[0008] A video acquisition module is used to collect video images on the expressway and within a preset range around the expressway at each monitoring point;

[0009] An image recognition module is used to recognize video images, extract characteristic information of hidden danger events based on the video images, and recognize hidden danger events associated with the inducement of pedestrians on the highway based on the characteristic information;

[0010] The event judgment module is used to judge whether the current hidden danger event meets the inducement of pedestrians on the highway based on the characteristic information when there is a hidden danger event. and the estimated time of occurrence ;

[0011] The warning module is used to determine the degree of compatibility of pedestrians with the inducement to enter the highway. , and the estimated time of occurrence , generate warning information, and send warning information to the potential danger event point and the rear vehicle respectively.

[0012] The principle and advantage of the present invention are: through the identification and analysis of historical data, the inducement of pedestrians on the highway is traced back. After that, the video images of the highway and the preset range around the highway collected at the monitoring point are used to obtain the real-time monitored video images, and the characteristic information in the video images is extracted. After the characteristic information is extracted, it is identified whether there are hidden danger events associated with the inducement of pedestrians on the highway on the current highway according to the characteristic information. When there is a hidden danger event associated with the pedestrian on the highway event identified by the characteristic information, it is identified that the current hidden danger event meets the pedestrian on the highway inducement. The higher the degree of fit, the greater the possibility of pedestrians on the highway. At the same time, the time of pedestrians on the highway is identified according to the characteristic information. Finally, according to the degree of fit and the expected time of occurrence, early warning information is generated, and the early warning information is delivered to the rear vehicles and hidden danger event points.

[0013] Compared with the prior art, in this application, by pre-judging the event of pedestrians going on the highway, the possible event of pedestrians going on the highway is identified through various characteristic information before the pedestrians go on the highway. The characteristic information for predicting the occurrence of pedestrians going on the highway is obtained by tracing back the historical events of pedestrians going on the highway, and the inducements for pedestrians going on the highway are obtained through a large amount of data learning and analysis, so as to judge whether the characteristic information in the real-time collected video surveillance screen has the inducement to cause the pedestrian going on the highway, and thus judge whether the pedestrian going on the highway is likely to occur. By pre-judging the pedestrians going on the highway in advance and predicting the time of pedestrians going on the highway, the pedestrians are warned in advance and the vehicles behind are informed, thereby reducing the probability of pedestrians going on the highway.

[0014] Furthermore, the historical data module includes a recognition feature module, a process duration module and an inducement training module;

[0015] A recognition feature module is used to extract recognition features of the pedestrian in the video image when the pedestrian first appears in the video image based on the historical data of pedestrians entering the highway, and the recognition features include environmental features, facility features and personnel action features;

[0016] The process duration module is used to identify the time node when the pedestrian first appears in the video image and the time node when the pedestrian enters the highway based on the video image, and obtain the process duration;

[0017] The inducement training module is used to package the identification features and process duration of a pedestrian entering the highway event to obtain an analysis data packet. Based on the analysis data packets of several pedestrian entering the highway events as training sets, the inducement features contained in different pedestrian entering the highway events and the induction time relationship of various inducement features are obtained through machine learning.

[0018] Through the historical data of pedestrians on the highway event, the video images are feature extracted, and the features include environmental features, facility features, and personnel action features. For example, the damage of the highway fence may cause people in the surrounding area to go on the highway; when a traffic accident occurs on the highway, the driver and passengers may go on the highway while waiting to be handled, and in bad weather conditions, people's judgment is reduced, and the situation of mistakenly going on the highway occurs. After that, by detecting "people", when a person is detected in the video image, it is marked as the time point when the pedestrian appears, and then combined with the time point of the pedestrian on the highway event, the process duration from the appearance of the pedestrian in the video surveillance to the pedestrian on the highway is obtained. The three features under different situations generate different process durations of pedestrian on the highway events. For example, after a traffic accident occurs on the highway, through the analysis of personnel actions, it is found that when the personnel show the characteristics of impatience, the possibility of pedestrians going on the highway is greater and the process duration is shorter. For example, when bad weather reduces personnel judgment, the possibility of pedestrians going on the highway is greater than in normal weather, and the process duration is shorter.

[0019] Afterwards, the regularities were summarized through machine learning to obtain the causal characteristics of different pedestrian-on-highway incidents and the induced-duration relationship of the causal characteristics.

[0020] Further, the inducement training module includes an inducement coefficient module and a duration coefficient module;

[0021] The inducement coefficient module is used to identify the influence coefficient of each inducement feature on the pedestrian high-speed incident according to the occurrence frequency of various inducement features in the analysis data packet of the pedestrian high-speed incident. ;

[0022] The duration coefficient module is used to analyze the induced duration relationship of each inducement feature according to the analysis data packet of the pedestrian high-speed incident containing at least one different inducement feature, wherein the induced duration relationship includes increase duration, decrease duration and no effect, and determine the influence coefficient of the induced duration relationship of the inducement feature on the process duration according to the process duration in the analysis data packet. .

[0023] The influence coefficient of the inducement feature on pedestrians going on the highway is determined by the frequency of occurrence of each inducement feature. The higher the frequency of occurrence, the greater the influence of the inducement feature on pedestrians going on the highway. For example, among a total of 100 pedestrians going on the highway, 50 of them include traffic accidents as inducement features, and 30 include bad weather as inducement features. The influence coefficient of traffic accidents as inducement features is higher than that of bad weather. At the same time, according to the analysis data packets of pedestrians going on the highway events containing at least one different inducement feature. For example, the inducement features included in one data packet include traffic accidents, bad weather, and impatient emotions of personnel, the inducement features included in the second data packet include traffic accidents and bad weather, and the inducement features included in the third data packet include traffic accidents, bad weather, and facility damage. The duration of the process of pedestrians going on the highway in the first data packet is 1 minute, the duration of the process of pedestrians going on the highway in the second data packet is 2 minutes, and the duration of the process of pedestrians going on the highway in the third data packet is 2 minutes. Therefore, it is determined that the impatient emotions of personnel have an impact on the induction time of pedestrians going on the highway, while the facility damage has no impact on the induction time of personnel going on the highway. The greater the influence coefficient of the causal characteristic, the longer it takes to affect the change.

[0024] Furthermore, the image recognition module includes a feature extraction module and a feature judgment module;

[0025] Feature extraction module, used to extract feature information from video images in real time;

[0026] The feature judgment module is used to judge whether the extracted feature information is a causal feature. If it is a causal feature, the causal feature type is determined and it is judged as a potential danger event.

[0027] By extracting feature information from the real-time collected video images, we can identify whether the feature information is a predisposing feature trained based on historical data. If a predisposing feature exists, it is determined that a hidden danger event exists.

[0028] Further, the event judgment module includes a fit judgment module;

[0029] The fit judgment module is used to judge the impact of the induced characteristics and the influence coefficient of the induced characteristics on the pedestrian on the highway when there is a hidden danger event. , to determine the degree of compatibility of the pedestrian's incentive to go on the highway ;

[0030]

[0031] When the fit When the pedestrian enters the highway, it is judged that a pedestrian will enter the highway. The larger the value, the greater the possibility of pedestrians entering the highway. is the matching threshold, which is the matching degree of the causal features obtained from the pedestrians on the highway events in the historical data. The number of causal features included in this hidden danger event.

[0032] When judging the compatibility of the inducement for pedestrians on the highway, after obtaining the inducement features in the image, the influence coefficient of the inducement features is then determined according to whether the sum of the influence coefficients of the included inducement features is greater than or equal to the preset compatibility threshold. When it is greater than or equal to, it is possible that a pedestrian may enter the highway, and the greater the degree of fit C, the greater the possibility of pedestrians entering the highway.

[0033] Furthermore, the event judgment module also includes a time judgment module

[0034] Time judgment module, when the fit When the influence coefficient of the induction time relationship of the current hidden danger characteristics on the process time is Get the influence weight of each current hidden danger feature on the induction duration , determine the estimated time of occurrence of pedestrians entering the highway :

[0035]

[0036] in The average basic time is obtained based on the induction time relationship of all the induction characteristics obtained from historical data, which is no effect. It is the average impact duration of the induced duration relationship. When the duration increases, the value is positive, and when the duration decreases, the value is negative.

[0037] When the fit is identified When the influence coefficient of the current incentive data on the process duration is Get the influence weight of each current hidden danger feature on the induction duration As mentioned above, for the influence coefficient , is determined by the length of the impact on the process duration. If there is no impact, it is 0. The greater the impact on the process duration, the greater the impact coefficient The larger the influence coefficient of the current incentive characteristics , determine the various influence coefficients The influence weight Then according to the formula The final estimated duration of the pedestrian on-highway incident is calculated. After that, the rear vehicles and pedestrians can be warned based on the duration of the process. Pedestrian warnings can be made through nearby warning devices. If there is no warning device nearby, drones can also be dispatched for warning. The speed of the rear vehicle is detected, and vehicles that will reach the potential incident point after the process duration are warned. Warnings can be made through traffic information boards on the road. By warning vehicles, pedestrians can be prevented from ignoring warning information or the warning information cannot be delivered to pedestrians in time. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 Schematic diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0039] The following is further described in detail through specific implementation methods:

[0040] The embodiment is basically as shown in the attached Figure 1 As shown:

[0041] The road pedestrian detection system based on computer vision is characterized by: comprising a server, the server comprising a historical data module, a video acquisition module, an image recognition module, an event judgment module and an early warning module;

[0042] The historical data module is used to collect the pedestrians on the highway events in the historical data, and trace the event process forward according to the time point of the pedestrians on the highway events, and analyze the causes of pedestrians on the highway;

[0043] A video acquisition module is used to collect video images on the expressway and within a preset range around the expressway at each monitoring point;

[0044] The image recognition module is used to recognize video images, extract characteristic information of hidden danger events based on the video images, and identify hidden danger events associated with the inducement of pedestrians on the highway based on the characteristic information.

[0045] Specifically, when performing feature extraction, micro-pedestrian features (such as wandering duration, sudden increase in moving speed) and macro-environmental features (such as broken surrounding fences, traffic accident scenes) are extracted simultaneously, and the prediction robustness is improved through feature cross-validation.

[0046] The event judgment module is used to judge whether the current hidden danger event meets the inducement of pedestrians on the highway based on the characteristic information when there is a hidden danger event. and the estimated time of occurrence ;

[0047] The warning module is used to determine the degree of compatibility of pedestrians with the inducement to enter the highway. , and the estimated time of occurrence , generate warning information, and send warning information to the potential danger event point and the rear vehicle respectively.

[0048] Through the identification and analysis of historical data, the causes of pedestrians on the highway are traced back. After that, the real-time monitored video images are obtained through the video images of the highway and the preset range around the highway collected at the monitoring point, and the feature information in the video images is extracted. After the feature information is extracted, it is identified whether there are hidden danger events associated with the inducement of pedestrians on the highway on the current highway. When there is a hidden danger event associated with the pedestrian on the highway event identified by the feature information, it is identified that the current hidden danger event meets the pedestrian on the highway inducement. The higher the degree of fit, the greater the possibility of pedestrians on the highway. At the same time, the time of pedestrians on the highway is identified according to the feature information. Finally, according to the fit and the expected time of occurrence, the warning information is generated, and the warning information is delivered to the rear vehicles and hidden danger event points.

[0049] By pre-judging the event of pedestrians entering the highway, various characteristic information can be used to identify possible pedestrians entering the highway before they enter the highway. The characteristic information for predicting the occurrence of pedestrians entering the highway is obtained by tracing back historical pedestrians entering the highway events, and the inducements for pedestrians to enter the highway are obtained through a large amount of data learning and analysis. In this way, the characteristic information in the real-time video surveillance screen is judged to see whether there are inducements for pedestrians to enter the highway, so as to judge whether pedestrians may enter the highway. By pre-judging the occurrence of pedestrians entering the highway and the time of pedestrians entering the highway, pedestrians are warned in advance and the vehicles behind are informed, thereby reducing the probability of pedestrians entering the highway.

[0050] The historical data module includes a recognition feature module, a process duration module and an inducement training module;

[0051] A recognition feature module is used to extract recognition features of the pedestrian in the video image when the pedestrian first appears in the video image based on the historical data of pedestrians entering the highway, and the recognition features include environmental features, facility features and personnel action features;

[0052] The process duration module is used to identify the time node when the pedestrian first appears in the video image and the time node when the pedestrian enters the highway based on the video image, and obtain the process duration;

[0053] The inducement training module is used to package the identification features and process duration of a pedestrian entering the highway event to obtain an analysis data packet. Based on the analysis data packets of several pedestrian entering the highway events as training sets, the inducement features contained in different pedestrian entering the highway events and the induction time relationship of various inducement features are obtained through machine learning.

[0054] Using an incremental learning framework, newly collected event data is added to the training set every day, the influence coefficient M of the inducement feature and the inducement duration coefficient N are dynamically adjusted, and the model iteration effect is verified through A / B testing.

[0055] Through the historical data of pedestrians on the highway, the video images are feature extracted, and the features include environmental features, facility features, and personnel action features. For example, the damage of the highway fence may cause people in the surrounding area to go on the highway; when a traffic accident occurs on the highway, the driver and passengers may go on the highway while waiting to be handled, and in bad weather conditions, people's judgment is reduced, and the situation of mistakenly going on the highway occurs. After that, by detecting "people", when a person is detected in the video image, it is marked as the time point when the pedestrian appears, and then combined with the time point of the pedestrian on the highway event, the process duration of the pedestrian on the highway from the video surveillance is obtained. The three features under different situations generate different process durations of pedestrian on the highway events. For example, after a traffic accident occurs on the highway, through the analysis of personnel actions, it is found that when the personnel show impatience, the possibility of pedestrians going on the highway is greater and the process duration is shorter. For example, when bad weather reduces personnel judgment, the possibility of pedestrians going on the highway is greater than in normal weather, and the process duration is shorter.

[0056] Specifically, for historical data, this application integrates highway surveillance videos, historical event databases of traffic management departments, vehicle camera data, and emergency reports from third-party platforms (such as traffic broadcasts and social media) to form a multi-dimensional data source, and then ensures data quality through data cleaning steps such as denoising, timestamp alignment, and outlier removal. For example, the system is connected to a provincial traffic management platform to obtain historical pedestrian incident data on highways across the country in real time, and privacy compliance is ensured through a data desensitization protocol.

[0057] After that, according to the number of pedestrian incidents on different sections, the monitoring equipment of subsequent highways will be supplemented as data support. The sections with more pedestrian incidents will be given priority for improving monitoring coverage. At the same time, according to the number of pedestrian incidents on each section, mobile monitoring equipment such as drones and patrol car cameras can be deployed in accident-prone areas, and high-frequency collection mode can be started during holidays or in bad weather to dynamically supplement key data.

[0058] At the same time, through the enhancement method of video data, such as using the adversarial network (GAN) to generate simulated scenes in different weather conditions (rainy days, foggy days, and heavy snowy days), and expanding pedestrian behavior fragments through the time difference method, the problem of insufficient data for rare events is solved. For example, the adversarial network (GAN) is used to simulate monitoring images in bad weather, and the pedestrian actions are time-series expanded to generate diversified training samples.

[0059] Afterwards, the regularities were summarized through machine learning to obtain the causal characteristics of different pedestrian-on-highway incidents and the induced-duration relationship of the causal characteristics.

[0060] The inducement training module includes an inducement coefficient module and a duration coefficient module;

[0061] The inducement coefficient module is used to identify the influence coefficient of each inducement feature on the pedestrian high-speed incident according to the occurrence frequency of various inducement features in the analysis data packet of the pedestrian high-speed incident. ;

[0062] The duration coefficient module is used to analyze the induced duration relationship of each inducement feature according to the analysis data packet of the pedestrian high-speed incident containing at least one different inducement feature, wherein the induced duration relationship includes increase duration, decrease duration and no effect, and determine the influence coefficient of the induced duration relationship of the inducement feature on the process duration according to the process duration in the analysis data packet. .

[0063] The influence coefficient of the inducement feature on pedestrians going on the highway is determined by the frequency of occurrence of each inducement feature. The higher the frequency of occurrence, the greater the influence of the inducement feature on pedestrians going on the highway. For example, among a total of 100 pedestrians going on the highway, 50 of them include traffic accidents as inducement features, and 30 include bad weather as inducement features. The influence coefficient of traffic accidents as inducement features is higher than that of bad weather. At the same time, according to the analysis data packets of pedestrians going on the highway events containing at least one different inducement feature. For example, the inducement features included in one data packet include traffic accidents, bad weather, and impatient emotions of personnel, the inducement features included in the second data packet include traffic accidents and bad weather, and the inducement features included in the third data packet include traffic accidents, bad weather, and facility damage. The duration of the process of pedestrians going on the highway in the first data packet is 1 minute, the duration of the process of pedestrians going on the highway in the second data packet is 2 minutes, and the duration of the process of pedestrians going on the highway in the third data packet is 2 minutes. Therefore, it is determined that the impatient emotions of personnel have an impact on the induction time of pedestrians going on the highway, while the facility damage has no impact on the induction time of personnel going on the highway. The larger the impact coefficient, the more time it takes to affect the change.

[0064] The image recognition module includes a feature extraction module and a feature judgment module;

[0065] Feature extraction module, used to extract feature information from video images in real time;

[0066] The feature judgment module is used to judge whether the extracted feature information is a causal feature. If it is a causal feature, the causal feature type is determined and the existence of a hidden danger event is judged.

[0067] By extracting feature information from the real-time collected video images, we can determine whether the feature information is a predisposing feature trained based on historical data. If a predisposing feature exists, it is determined that a potential danger event exists.

[0068] The event judgment module includes a fit judgment module;

[0069] The fit judgment module is used to judge the impact of the induced characteristics and the influence coefficient of the induced characteristics on the pedestrian on the highway when there is a hidden danger event. , to determine the degree of compatibility of the pedestrian's incentive to go on the highway ;

[0070]

[0071] When the fit When the pedestrian enters the highway, it is judged that a pedestrian will enter the highway. The larger the value, the greater the possibility of pedestrians entering the highway. is the matching threshold, which is the matching degree of the causal features obtained from the pedestrians on the highway events in the historical data. The number of causal features included in this hidden danger event.

[0072] When judging the compatibility of the inducement for pedestrians on the highway, after obtaining the inducement features in the image, the influence coefficient of the inducement features is then determined according to whether the sum of the influence coefficients of the included inducement features is greater than or equal to the preset compatibility threshold. When it is greater than or equal to, it is possible that a pedestrian may enter the highway, and the greater the degree of fit C, the greater the possibility of pedestrians entering the highway.

[0073] The event judgment module also includes a time judgment module

[0074] Time judgment module, when the fit When the influence coefficient of the induction time relationship of the current hidden danger characteristics on the process time is Get the influence weight of each current hidden danger feature on the induction duration , determine the estimated time of occurrence of pedestrians entering the highway :

[0075]

[0076] in The average basic time is obtained based on the induction time relationship of all the induction characteristics obtained from historical data, which is no effect. It is the average impact duration of the induced duration relationship. When the duration increases, the value is positive, and when the duration decreases, the value is negative.

[0077] When the identification is obtained When the influence coefficient of the current incentive data on the process duration is Get the influence weight of each current hidden danger feature on the induction duration As mentioned above, for the influence coefficient , is determined by the length of the impact on the process duration. If there is no impact, it is 0. The greater the impact on the process duration, the greater the impact coefficient The larger the influence coefficient of the current incentive characteristics , determine the various influence coefficients The influence weight Then according to the formula The final estimated duration of the pedestrian incident on the highway is calculated. After that, the rear vehicles and pedestrians can be warned according to the duration of the process. Pedestrian warnings can be carried out through nearby warning devices. If there is no warning device nearby, drones can also be dispatched for warning. The speed of the rear vehicle is detected, and the vehicle that will reach the potential event point after the process duration is warned. The warning can be issued through the traffic information board on the road. By warning the vehicle, pedestrians can be prevented from ignoring the warning information or the warning information cannot be delivered to the pedestrians in time.

[0078] The recognition feature module also uses a pre-trained behavioral psychology model (LSTM+Attention) to identify abnormal emotional actions of pedestrians (such as waving quickly, turning back and forth repeatedly, etc.), determine whether pedestrians have abnormal emotions, and quantify the probability of triggering irrational behaviors. And according to the probability of triggering irrational behaviors, the degree of fit is evaluated. Make adjustments.

[0079]

[0080] P is the probability of triggering irrational behavior. , make adjustments to improve the trigger sensitivity when pedestrians show abnormal emotions.

[0081] At the same time, when the fit After the adjustment is made, the event judgment module will Approaching Threshold When a traffic signal is detected, the manual review interface is automatically triggered, and the video clips and feature analysis results are pushed to the traffic control center, where the on-duty personnel will judge and execute the prediction strategy.

[0082] The above are only embodiments of the present invention. Common knowledge such as the known specific structures and characteristics in the scheme is not described in detail here. Ordinary technicians in the relevant field know all the common technical knowledge in the technical field to which the invention belongs before the application date or priority date, can obtain all the existing technologies in the field, and have the ability to apply conventional experimental means before that date. Ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the scope of protection of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A road pedestrian detection system based on computer vision, characterized by: The server includes a historical data module, a video acquisition module, an image recognition module, an event judgment module and an early warning module; The historical data module is used to collect the pedestrians on the highway events in the historical data, and trace the event process forward according to the time point of the pedestrians on the highway events, and analyze the causes of pedestrians on the highway; A video acquisition module is used to collect video images on the expressway and within a preset range around the expressway at each monitoring point; An image recognition module is used to recognize video images, extract characteristic information of hidden danger events based on the video images, and recognize hidden danger events associated with the inducement of pedestrians on the highway based on the characteristic information; The event judgment module is used to judge whether the current hidden danger event meets the inducement of pedestrians on the highway based on the characteristic information when there is a hidden danger event. and the estimated time of occurrence ; The warning module is used to determine the degree of compatibility of pedestrians with the inducement to enter the highway. , and the estimated time of occurrence , generate warning information, and send warning information to the potential danger event point and the rear vehicle respectively; The historical data module includes a recognition feature module, a process duration module and an inducement training module; A recognition feature module is used to extract recognition features of the pedestrian in the video image when the pedestrian first appears in the video image based on the historical data of pedestrians entering the highway, and the recognition features include environmental features, facility features and personnel action features; The process duration module is used to identify the time node when the pedestrian first appears in the video image and the time node when the pedestrian enters the highway based on the video image, and obtain the process duration; The inducement training module is used to package the identification features and process duration of a pedestrian on-highway event to obtain an analysis data package. Based on the analysis data packages of several pedestrian on-highway events as training sets, the inducement features contained in different pedestrian on-highway events and the induction duration relationship of various inducement features are obtained through machine learning; The inducement training module includes an inducement coefficient module and a duration coefficient module; The inducement coefficient module is used to identify the influence coefficient of each inducement feature on the pedestrian high-speed incident according to the occurrence frequency of various inducement features in the analysis data packet of the pedestrian high-speed incident. ; The duration coefficient module is used to analyze the induced duration relationship of each inducement feature according to the analysis data packet of the pedestrian high-speed incident containing at least one different inducement feature, wherein the induced duration relationship includes increase duration, decrease duration and no effect, and determine the influence coefficient of the induced duration relationship of the inducement feature on the process duration according to the process duration in the analysis data packet. ; The image recognition module includes a feature extraction module and a feature judgment module; Feature extraction module, used to extract feature information from video images in real time; A feature judgment module is used to judge whether the extracted feature information is a causal feature. If it is a causal feature, the causal feature type is determined and it is judged as a potential danger event. The event judgment module includes a fit judgment module; The fit judgment module is used to judge the impact of the induced characteristics and the influence coefficient of the induced characteristics on the pedestrian on the highway when there is a hidden danger event. , to determine the degree of compatibility of the pedestrian's incentive to go on the highway ; When the fit When the pedestrian enters the highway, it is judged that a pedestrian will enter the highway. The larger the value, the greater the possibility of pedestrians entering the highway. is the matching threshold, which is the matching degree of the causal features obtained from the pedestrians on the highway events in the historical data. The number of causal features included in this hidden danger event; The event judgment module also includes a time judgment module; Time judgment module, when the fit When the influence coefficient of the induction time relationship of the current hidden danger characteristics on the process time is Get the influence weight of each current hidden danger feature on the induction duration , determine the estimated time of occurrence of pedestrians entering the highway : in The average basic time for all the induced characteristics with no effect on the induced duration relationship obtained based on historical data, It is the average impact duration of each induced factor characteristic that has an impact on the induced duration relationship. When the duration increases, the value is positive, and when the duration decreases, the value is negative.

Images