Traffic incident secondary identification method and system based on deep learning

Abstract

The invention relates to the technical field of intelligent traffic, in particular to a traffic incident secondary recognition method and system based on deep learning, and the method comprises the steps: monitoring and collecting traffic information in real time, reporting the traffic information when the traffic information is recognized as a traffic incident, processing the traffic incident information, and forming a database; an event target and an event type are confirmed, the event target is confirmed by using a secondary identification model, and if the event target does not accord with the event type, false alarm reminding is executed; for the traffic events, whether report records with the same event type exist in a database within unit time is retrieved, if not, normal report is executed, the report records of the database are updated, and if yes, repeated traffic event report filtering is executed; classifying the event types of the traffic events, and merging the multiple levels of traffic events; judging whether the event type of the traffic event is matched with a related business rule or not, and if yes, filtering the traffic event; and the safety, the stability and the working efficiency of the monitoring system are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to a secondary recognition method and system for traffic events based on deep learning. Background Art

[0002] At present, traffic incident detection technology based on cameras or radar sensors is becoming more and more mature, and the individual detection effect from the algorithm perspective is getting better and better. Relevant departments have formulated and announced multiple standards for video traffic incident detectors, stipulating the definitions of events such as stopping, driving in the wrong direction, making phone calls, and congestion in road traffic. After years of practical application, the following two problems exist.

[0003] First, due to the lack of uniformity in the definition of event targets in the current standards, some of the same event targets are identified as different event targets under different cameras. For example, the same problem vehicle with a malfunction, violation or other abnormal condition will issue an early warning under different cameras. Under normal circumstances, only one early warning is required when it is first discovered. However, since the standard does not clearly specify the identification method and tracking mechanism for problem vehicles, some video traffic event detectors are unable to achieve continuous identification and tracking of problem vehicles under different cameras, resulting in information duplication and false alarms.

[0004] Secondly, due to the incomplete definition of event types in the current standards, some situations determined as events by the algorithm do not meet actual business needs. For example, a parking event determined by the algorithm may actually be maintenance work carried out by the road management department. Maintenance refers to the repair and cleaning of roads, which generally does not require early warning or special treatment. However, since the standard does not clearly define the distinction and treatment methods for maintenance and parking events, some video traffic event detectors are unable to recognize the difference between maintenance and parking events, resulting in information errors and interference.

[0005] The information disclosed in this background technology section is only intended to deepen the understanding of the overall background technology of the present invention, and should not be regarded as acknowledging or suggesting in any form that the information constitutes the prior art known to those skilled in the art. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a method and system for secondary identification of traffic events based on deep learning, to perform secondary confirmation on the reported information of traffic events, to screen out repeated events, to merge information of multiple traffic events, and to display the whole process of traffic event processing.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is: a secondary identification method of traffic events based on deep learning, comprising the following steps:

[0008] Real-time monitoring and collection of traffic information, and reporting of the traffic incident when it is identified as a traffic incident, and data storage and processing of the information of the traffic incident to form a database;

[0009] Confirming the event target and event type of the traffic event, confirming the event target using a secondary recognition model, and executing a false alarm reminder if the event target of the traffic event does not match the actual event target;

[0010] For the traffic event, search the database to see whether there is a report record with the same event type as the traffic event within a unit time. If not, perform normal reporting and update the database report record. If so, perform duplicate traffic event report filtering.

[0011] Classifying the event types of the traffic events and merging multiple levels of the traffic events;

[0012] It is determined whether the event type of the traffic event matches the relevant business rules. If it matches, the traffic event is filtered.

[0013] Furthermore, a detection perception front end is set to monitor and collect traffic information in real time, and the traffic information includes image appearance features, image acquisition time and the pile number of the detection perception front end.

[0014] Furthermore, the detection and perception front end may be configured as a video detector; and the event targets include persons and vehicles.

[0015] Further, the event target and event type of the traffic event are confirmed, and the event target is confirmed using a secondary recognition model. If the event target of the traffic event does not match the actual event target, a false alarm reminder is performed, including the following steps:

[0016] Extracting the event target and event type of the traffic event according to the reported traffic event;

[0017] According to the event type of the traffic event, selecting a corresponding secondary recognition model;

[0018] Extracting the event target of the traffic event to form a corresponding ROI image, and inputting the ROI image into the secondary recognition model for false alarm judgment and confirmation;

[0019] If and only if the secondary recognition model confirms that the event target of the traffic event corresponding to the ROI image does not match the actual event target, the traffic event is judged to be a false alarm, and a false alarm reminder is executed.

[0020] Furthermore, for the traffic event, searching the database for whether there is a report record of the same event type as the traffic event within a unit time, if not, performing normal reporting and updating the database report record, if yes, performing duplicate traffic event report filtering, including the following steps:

[0021] While reporting the traffic event, confirming that the pile number corresponding to the detection perception front end that reports the traffic event is a key value;

[0022] Check the database to determine whether there is a report record with the same event type as the traffic event under the pile number corresponding to the detection perception front end, and if so, compare the traffic time report time with the latest report time recorded in the database to determine the time interval;

[0023] If the time interval is less than the time threshold, it is determined whether the currently reported traffic event is the same event as the most recently reported traffic event in the database. If so, duplicate traffic event reporting filtering is performed; if not, normal reporting is performed and the database reporting record is updated.

[0024] Further, the method comprises the following steps: determining whether the currently reported traffic event is the same event as the most recently reported traffic event in the database; if so, performing duplicate traffic event reporting filtering; if not, performing normal reporting and updating the database reporting record; and

[0025] Extract the appearance feature vector corresponding to the ROI image according to the event target of the traffic event confirmed by the secondary recognition model

[0026] Retrieve and extract the appearance feature vector of the ROI image corresponding to the traffic event recently reported in the database

[0027] Determine the appearance feature vector of the event target of the currently reported traffic event The event target appearance feature vector of the traffic incident recently reported in the database The minimum cosine distance Cos d ;

[0028] If the minimum cosine distance Cos d If the distance is greater than the distance threshold, it is determined that the traffic event currently reported in the database record is the same as the traffic event most recently reported in the database, and duplicate traffic event reporting filtering is performed; otherwise, normal reporting is performed and the database reporting record is updated.

[0029] Furthermore, the minimum cosine distance Cos d Obtained by the following formula:

[0030]

[0031] in, represents the appearance feature vector of the event target of the currently reported traffic event, represents the appearance feature vector of the event target of the traffic event recently reported in the database, represents the dot product of the appearance feature vector of the event target of the currently reported traffic event and the appearance feature vector of the event target of the most recently reported traffic event in the database, and Represent the appearance feature vectors and The norm of .

[0032] Furthermore, the event types of the traffic events are classified and multiple levels of traffic events are merged and processed, including the following steps:

[0033] When the event target of the traffic incident is a person, the secondary recognition model classifies the event target of the traffic incident into social personnel, maintenance personnel or law enforcement personnel according to the functions of the personnel, and supplements the information of the event target;

[0034] When the event target of the traffic incident is a vehicle, the secondary recognition model classifies the event target of the traffic incident into social vehicles, maintenance vehicles, law enforcement vehicles or hazardous chemicals vehicles according to the functions of the vehicles, and supplements the information of the event target;

[0035] Merging multi-level traffic events according to the event target information of the traffic events by the secondary recognition model, wherein the multi-level traffic events include the same type of traffic events reported continuously in the spatiotemporal dimension and multiple types of traffic events reported continuously under a single pile number;

[0036] The development process of the multi-level traffic event is displayed under the merged multi-level traffic event, and the progress snapshot photos and change process tracking videos of the multi-level traffic event are retained and recorded.

[0037] Further, the determining whether the event type of the traffic event matches the relevant business rules, and if so, filtering the traffic event, comprises the following steps:

[0038] Based on the event type of the traffic event determined by the secondary recognition model, determine all filtering rules corresponding to the event type;

[0039] It is determined whether the event type of the traffic event belongs to any one of the filtering rules. If it matches, the traffic event is filtered. If it does not match, the traffic event is reported and confirmed for a second time.

[0040] The present invention also provides a traffic event secondary recognition system based on deep learning, comprising:

[0041] The data collection module is provided with a detection and perception front end for real-time monitoring and collection of traffic information, and reporting the traffic incident when it is identified as a traffic incident;

[0042] A data processing module, provided with a microprocessor and an embedded neural network processor, for receiving information of the traffic event and performing data processing;

[0043] A data storage module, used to receive and store traffic event information to form an object storage database and a traffic event storage database;

[0044] A data transmission module, used to realize communication transmission among the data acquisition module, the data processing module, the data storage module and the data center;

[0045] The data processing module includes a traffic incident false alarm reminder unit, a repeated traffic incident reporting filtering unit, a multi-level traffic incident merging processing unit, and a traffic incident confirmation unit based on business logic;

[0046] The traffic incident false alarm reminder unit is used to confirm the event target and event type of the traffic incident, and to confirm the event target using a secondary recognition model. If the event target of the traffic incident does not match the actual event target, a false alarm reminder is executed;

[0047] The repeated traffic event reporting filtering unit is used to search the database for a traffic event to see if there is a report record with the same event type as the traffic event within a unit time. If not, normal reporting is performed and the database report record is updated. If so, repeated traffic event reporting filtering is performed.

[0048] The multi-level traffic event merging processing unit is used to classify the event types of the traffic events and merge the multi-level traffic events;

[0049] The traffic event confirmation unit based on business logic is used to determine whether the event type of the traffic event matches the relevant business rules. If it matches, the traffic event is filtered.

[0050] The beneficial effects of the present invention are as follows: the present invention confirms reported traffic events by setting a secondary recognition model for event targets, and can reduce the cost of human review and improve the detection accuracy of the event detection platform for the problems of misidentification and false reporting in the traffic event detection system, thereby helping to improve the safety and stability of the traffic event detection system; and adopts a method of filtering repeatedly reported traffic events, and can reduce data redundancy, improve data readability and analysis efficiency, and reduce labor costs and improve overall work efficiency for the problems of frequent traffic event reporting and repeated reporting of the same event target under the same pile number, while reducing labor costs and improving overall work efficiency; and through a method of merging multiple levels of traffic events and coordinating traffic event confirmation processing with business logic, it can introduce a custom business logic function for the problem of large differences between the traffic event reporting of the road network management platform fed back by manual telephone and the traffic event detection reporting data based on machine recognition, thereby improving user work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0052] Figure 1 Schematic diagram of the steps of a secondary identification method for traffic events based on deep learning in an embodiment of the present invention;

[0053] Figure 2 A schematic diagram of the steps of reminding of false alarm of traffic incidents in an embodiment of the present invention;

[0054] Figure 3 A schematic diagram of the processing process of a traffic incident false alarm reminder example in an embodiment of the present invention;

[0055] Figure 4 A schematic diagram of the steps of repeatedly filtering reported traffic events in an embodiment of the present invention;

[0056] Figure 5 A schematic diagram of the steps for judging a traffic event in an embodiment of the present invention;

[0057] Figure 6 This is a flow chart of filtering of repeatedly reported traffic events in an embodiment of the present invention;

[0058] Figure 7 A schematic diagram of a processing process of a repeated traffic event filtering example in an embodiment of the present invention;

[0059] Figure 8 A schematic diagram of the steps of merging and processing multi-level traffic events in an embodiment of the present invention;

[0060] Fig. 9 A schematic diagram of the steps of traffic event confirmation based on business logic in an embodiment of the present invention;

[0061] Fig.10 Schematic diagram of the structure of a secondary traffic event recognition system based on deep learning in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0063] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only and do not represent the only implementation method.

[0064] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0065] like Figures 1 to 9 The traffic incident secondary recognition method based on deep learning shown includes the following steps:

[0066] Real-time monitoring and collection of traffic information, reporting of traffic incidents when identified as such, data storage and processing of traffic incident information to form a database;

[0067] Confirm the event target and event type of the traffic incident, and use the secondary recognition model to confirm the event target. If the event target of the traffic incident does not match the actual event target, a false alarm reminder will be executed;

[0068] For traffic incidents, search the database to see if there is a report record with the same event type as the traffic incident within a unit time. If not, perform normal reporting and update the database report record. If so, perform duplicate traffic incident report filtering.

[0069] Classify the types of traffic incidents and merge multi-level traffic incidents;

[0070] Determine whether the event type of the traffic event matches the relevant business rules. If it matches, filter the traffic event.

[0071] The present invention confirms reported traffic events by setting a secondary recognition model for event targets. In view of the problems of misidentification and false reporting in the traffic event detection system, it can reduce the cost of human review, improve the detection accuracy of the event detection platform, and help improve the safety and stability of the traffic event detection system; and adopts a method of filtering repeatedly reported traffic events. In view of the problems of frequent reporting of traffic events and repeated reporting of the same event target under the same pile number, it can reduce data redundancy, improve data readability and analysis efficiency, and reduce labor costs and improve overall work efficiency; through the method of multi-level traffic event merging and traffic event confirmation processing with business logic, in view of the large difference between the traffic event reporting situation of the road network management platform fed back by manual telephone and the traffic event detection reporting data based on machine recognition, it can introduce a custom business logic function, thereby improving user work efficiency.

[0072] It should be noted that the main reason for the large difference between the traffic incident reporting situation of the road network management platform based on manual telephone feedback and the traffic incident detection and reporting data based on machine recognition is the conflict between the algorithm definition and the business definition of traffic incidents. Specifically, the traffic incidents defined by the algorithm follow the logic of all-out inspection, but in actual business, traffic police officers are not particularly concerned about some traffic incidents, and traffic incidents are frequently reported, resulting in high user review costs. In order to improve system availability, multiple types of business logic are introduced to conduct secondary confirmation of traffic incident reports. For traffic incidents that are not of particular concern to traffic police officers, a strategy of recording reminders but not reporting is adopted. In addition, custom business logic functions are introduced to improve user work efficiency.

[0073] Based on the above embodiment, a detection perception front end is set to monitor and collect traffic information in real time. The traffic information includes image appearance features, image acquisition time and the pile number of the detection perception front end; the detection perception front end can be set as a video detector; event targets include people and vehicles.

[0074] By setting up a detection and perception front end, it is possible to more accurately identify event targets such as vehicles, personnel, and traffic signs, thereby providing data support for the subsequent data processing process. During detection, a video detector is used as the detection and perception front end, and visual recognition, data analysis, and real-time monitoring technology are used to collect traffic information such as pedestrians and vehicles in the field of view, and perform data analysis. Among them, by recording the pile number of the detection and perception front end as an identifier of the road position, the specific location of the traffic incident can be accurately located to facilitate subsequent rapid response and processing; by recording image acquisition events, it is helpful to analyze the distribution characteristics of traffic events in the subsequent events. Combined with event information and pile number information, traffic events at a specific time and place can be quickly retrieved and analyzed, thereby improving data processing efficiency.

[0075] On the basis of the above embodiment, the event target and event type of the traffic event are confirmed, and the event target is confirmed using a secondary recognition model. If the event target of the traffic event does not match the actual event target, a false alarm reminder is performed, including the following steps:

[0076] Extract the event target and event type of the traffic event according to the reported traffic event;

[0077] According to the event type of the traffic event, the corresponding secondary recognition model is selected;

[0078] The event target of the traffic event is extracted to form a corresponding ROI image, and the ROI image is input into the secondary recognition model for false alarm judgment and confirmation;

[0079] If and only if the secondary recognition model confirms that the event target of the traffic event corresponding to the ROI image does not match the actual event target, the traffic event is judged to be a false alarm and a false alarm reminder is executed.

[0080] Through automated video analysis, a secondary recognition model for event targets is introduced to confirm reported traffic events in order to address the problems of misidentification and false alarm in traffic event detection systems. This reduces reliance on manual monitoring, thereby reducing human review costs and improving the detection accuracy and efficiency of the event detection platform. This helps to improve the safety and stability of traffic event detection systems and further reduce the manpower, material and time costs incurred by erroneous event handling.

[0081] It should be noted that the secondary recognition model can form targeted analysis and enhanced recognition for different event types, such as pedestrian crossing or social vehicle crossing, so as to determine whether the reported traffic incident is consistent with the actual event target.

[0082] by Figure 3Take the traffic incident false alarm reminder event as an example. The video detector with the stake number AK2103+750 issued a traffic incident report and detected a pedestrian intruding. At this time, the event target was extracted and the appearance features were stored. The secondary recognition model corresponding to the pedestrian was selected for secondary confirmation. The secondary recognition model judged that the event target was a roadblock, and its appearance features did not match those of a pedestrian. The traffic incident was judged to be a system false alarm, and a false alarm reminder was executed.

[0083] On the basis of the above embodiment, for traffic events, whether there is a report record with the same event type as the traffic event in the database within a unit time, if not, then perform normal reporting and update the database report record, if yes, then perform duplicate traffic event report filtering, including the following steps:

[0084] When reporting a traffic incident, confirm that the pile number corresponding to the detection perception front end of the reported traffic incident is the key value;

[0085] Check the database to determine whether there is a report record with the same event type as the traffic event under the pile number corresponding to the detection perception front end. If so, compare the traffic time report time with the latest report time recorded in the database to determine the time interval;

[0086] If the time interval is less than the time threshold, it is determined whether the currently reported traffic event is the same as the most recently reported traffic event in the database. If so, duplicate traffic event reporting filtering is performed; if not, normal reporting is performed and the database reporting record is updated.

[0087] By filtering duplicate traffic incident reports, the amount of information that the traffic management department needs to process is reduced, so that duplicate events can be quickly identified and filtered, data processing can be accelerated, and misjudgment and unnecessary responses caused by multiple reports of the same traffic incident can be avoided. This reduces data redundancy, improves data readability and analysis efficiency, and reduces labor costs while improving overall work efficiency. By screening the time threshold, it ensures that multiple detections of the traffic time are formed within the time period, prevents the occurrence of underreporting, and ensures the accuracy of the system processing results.

[0088] On the basis of the above embodiment, it is determined whether the currently reported traffic event is the same event as the traffic event recently reported in the database. If so, duplicate traffic event reporting filtering is performed; if not, normal reporting is performed and the database reporting record is updated; the following steps are included:

[0089] According to the event target of the traffic incident confirmed by the secondary recognition model, the appearance feature vector of the corresponding ROI image is extracted

[0091] Retrieve and extract the appearance feature vector of the ROI image corresponding to the most recently reported traffic incident in the database

[0092] Determine the appearance feature vector of the event target of the current reported traffic event The event target appearance feature vector of the most recently reported traffic incident in the database The minimum cosine distance Cos d ;

[0093] If the minimum cosine distance Cos d If the distance is greater than the distance threshold, the traffic event currently reported in the database record is determined to be the same as the traffic event most recently reported in the database, and duplicate traffic event reporting filtering is performed; otherwise, normal reporting is performed and the database reporting record is updated.

[0094] By calculating the appearance feature vector and the minimum cosine distance, the similarity between two event targets can be accurately compared, thereby improving the judgment accuracy. By comparing the minimum cosine distance between the two appearance feature vectors with the distance threshold, it is determined whether the two are the same event, reducing the occurrence of misjudgment. After the judgment, the database is updated in real time, which optimizes the maintenance and management of the database and ensures the timeliness of the data.

[0095] Based on the above embodiment, the minimum cosine distance Cos d Obtained by the following formula:

[0096]

[0097] in, Represents the event target appearance feature vector of the currently reported traffic event, Represents the event target appearance feature vector of the most recently reported traffic incident in the database, represents the dot product of the appearance feature vector of the event target of the currently reported traffic incident and the appearance feature vector of the event target of the most recently reported traffic incident in the database, and Represent the appearance feature vectors and The norm of .

[0098] Reference Figure 6The flowchart of repeated traffic event filtering is as follows: the time threshold is set to one hour and the distance threshold is set to 0.8. When reporting a traffic event, the secondary recognition model extracts appearance features and re-determines the event target and event type of the traffic event. The appearance features corresponding to the event target are retrieved from the database. If not, the event is directly judged, reported and processed, and the database is updated. If the same type of traffic event data exists under the pile number, the time interval between the two traffic events is checked. If the time interval is greater than the time threshold, it is reported and the database is updated. If it is less than the time threshold, the minimum cosine distance of the appearance feature vectors of the two traffic events is continued to be judged. If the minimum cosine distance is less than the distance threshold, it is reported and the database is updated. If it is greater than the distance threshold, the reporting filtering is performed, and the traffic event is not reported, and the database is updated at the same time.

[0099] by Figure 7 The repeated reporting of traffic event filtering events shown is an example. Two abnormal parking traffic events are reported successively in the same pile number, and the time interval is less than the time threshold. At this time, the appearance features corresponding to the target of the second traffic event are extracted, and the judgment is made through the appearance feature vector and the minimum cosine distance. If the similarity of the appearance of the time target of the two traffic times reaches the distance threshold, no reporting is performed. If it is less than the distance threshold, the reporting operation is performed.

[0100] Based on the above embodiment, the event types of traffic events are classified, and multi-level traffic events are merged and processed, including the following steps:

[0101] When the event target of a traffic incident is a person, the secondary recognition model divides the event target of the traffic incident into social personnel, maintenance personnel or law enforcement personnel according to the personnel's functions, and supplements the information of the event target;

[0102] When the event target of a traffic incident is a vehicle, the secondary recognition model divides the event target of the traffic incident into social vehicles, maintenance vehicles, law enforcement vehicles or hazardous chemicals vehicles according to the function of the vehicle, and supplements the information of the event target;

[0103] Based on the event target information of traffic events by the secondary recognition model, multi-level traffic events are merged. Multi-level traffic events include the same type of traffic events reported continuously in the time and space dimensions and multiple types of traffic events reported continuously under a single pile number.

[0104] The development process of the multi-level traffic events is displayed under the merged multi-level traffic events, and the progress snapshot photos and change process tracking videos of the multi-level traffic events are retained and recorded.

[0105] By subdividing the functions of event targets through the secondary recognition model, the types of traffic events can be distinguished more accurately, thereby supplementing the event targets and ensuring the completeness and thoroughness of event records, which will help subsequent traffic management departments to allocate resources more effectively to deal with different types of traffic events. By merging the same or multiple types of traffic events, it can respond to complex traffic conditions more quickly, reduce processing time, simplify the analysis of multiple events, reduce the complexity of data processing, and improve data processing efficiency; by recording the progress of multi-level traffic events, capturing photos and changing videos, the system's ability to track the development process of events is enhanced, thereby determining the overall development process of traffic events, providing traffic management personnel with richer information, and supporting more accurate decision-making.

[0106] On the basis of the above embodiment, it is determined whether the event type of the traffic event matches the relevant business rules. If it matches, the traffic event is filtered, including the following steps:

[0107] Based on the event type of the traffic event determined by the secondary recognition model, all filtering rules corresponding to the event type are determined;

[0108] Determine whether the event type of the traffic event belongs to any of the filtering rules. If it matches, filter the traffic event. If it does not match, report the traffic event for a second time for confirmation.

[0109] Through automated rule matching and filtering, business processes can be optimized, manual operations can be reduced, the level of automation in business execution can be improved, and traffic incidents that meet relevant business rules can be filtered out instead of being reported, thereby further improving the pertinence and efficiency of traffic incident handling.

[0110] Please refer to Table 1 below for a list of relevant business logic, which usually include: abnormal parking type non-key attention events include parking in the emergency lane for less than the defined time, construction vehicles stopping, being in a congested state, being in a construction area, the same vehicle staying for a long time, and parking in the square of a toll station; pedestrian intrusion type non-key attention events include pedestrians getting off the emergency lane briefly, being in a congested state, and being in a construction area; road construction, non-motor vehicle intrusion, traffic congestion and other non-key attention events are the merger of multiple pile number events; vehicle reversing non-key attention events include being in a construction area, the merger of multiple pile number events, etc., which are not listed here one by one.

[0111] By organizing the relevant business rules and defining the business logic rule matrix, combined with the secondary identification information, a comprehensive judgment is made on the event report, and reports of non-concern traffic events are filtered out; after additional new business rules are set up, the business rules can also be added to the business logic rule matrix at a later stage to enrich the screening conditions.

[0112] Table 1: List of event-related filtering rules

[0113]

[0114]

[0115] like Fig.10 As shown, the present invention also provides a secondary traffic event recognition system based on deep learning, comprising:

[0116] The data collection module is provided with a detection and perception front end for real-time monitoring and collection of traffic information, and reporting traffic events when identified as traffic events;

[0117] A data processing module, provided with a microprocessor and an embedded neural network processor, for receiving information about traffic events and performing data processing;

[0118] A data storage module, used to receive and store traffic event information to form an object storage database and a traffic event storage database;

[0119] Data transmission module, used to realize communication transmission between data acquisition module, data processing module, data storage module and data center;

[0120] The data processing module includes a traffic incident false alarm reminder unit, a repeated traffic incident reporting filtering unit, a multi-level traffic incident merging processing unit, and a traffic incident confirmation unit based on business logic;

[0121] The traffic incident false alarm reminder unit is used to confirm the event target and event type of the traffic incident, and use the secondary recognition model to confirm the event target. If the event target of the traffic incident does not match the actual event target, a false alarm reminder is executed;

[0122] The repeated traffic event filtering unit is used to search the database for a traffic event to see if there is a report record with the same event type as the traffic event within a unit time. If not, normal reporting is performed and the database report record is updated. If so, repeated traffic event reporting filtering is performed.

[0123] The multi-level traffic event merging processing unit is used to classify the event types of traffic events and merge and process the multi-level traffic events;

[0124] The traffic event confirmation unit based on business logic is used to determine whether the event type of the traffic event matches the relevant business rules. If it matches, the traffic event is filtered.

[0125] The specific working method of the above-mentioned traffic event secondary identification system has been explained in the above-mentioned embodiment and will not be repeated here.

[0126] Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A secondary identification method for traffic incidents based on deep learning, characterized in that: The following steps are involved: Real-time monitoring and collection of traffic information, and reporting of the traffic incident when it is identified as a traffic incident, and data storage and processing of the information of the traffic incident to form a database; Confirming the event target and event type of the traffic event, confirming the event target using a secondary recognition model, and executing a false alarm reminder if the event target of the traffic event does not match the actual event target; For the traffic event, search the database to see whether there is a report record with the same event type as the traffic event within a unit time. If not, perform normal reporting and update the database report record. If so, perform duplicate traffic event report filtering. Classifying the event types of the traffic events and merging multiple levels of the traffic events; It is determined whether the event type of the traffic event matches the relevant business rules. If it matches, the traffic event is filtered.

2. The method for secondary identification of traffic events based on deep learning according to claim 1 is characterized in that: A detection perception front end is set to monitor and collect traffic information in real time, where the traffic information includes image appearance features, image collection time, and the pile number of the detection perception front end.

3. The method for secondary identification of traffic events based on deep learning according to claim 2 is characterized in that: The detection perception front end can be set as a video detector; the event targets include people and vehicles.

4. The method for secondary identification of traffic events based on deep learning according to claim 3 is characterized in that: The confirming of the event target and event type of the traffic event, confirming the event target using a secondary recognition model, and executing a false alarm reminder if the event target of the traffic event does not match the actual event target, includes the following steps: Extracting the event target and event type of the traffic event according to the reported traffic event; According to the event type of the traffic event, selecting a corresponding secondary recognition model; Extracting the event target of the traffic event to form a corresponding ROI image, and inputting the ROI image into the secondary recognition model for false alarm judgment and confirmation; If and only if the secondary recognition model confirms that the event target of the traffic event corresponding to the ROI image does not match the actual event target, the traffic event is judged to be a false alarm, and a false alarm reminder is executed.

5. The method for secondary identification of traffic events based on deep learning according to claim 3 is characterized in that: The method of searching the database for a traffic event to see whether there is a report record with the same event type as the traffic event within a unit time, and if not, performing normal reporting and updating the database report record, and if so, performing duplicate traffic event report filtering, includes the following steps: While reporting the traffic event, confirming that the pile number corresponding to the detection perception front end that reports the traffic event is a key value; Check the database to determine whether there is a report record with the same event type as the traffic event under the pile number corresponding to the detection perception front end, and if so, compare the traffic time report time with the latest report time recorded in the database to determine the time interval; If the time interval is less than the time threshold, it is determined whether the currently reported traffic event is the same event as the most recently reported traffic event in the database. If so, duplicate traffic event reporting filtering is performed; if not, normal reporting is performed and the database reporting record is updated.

6. The method for secondary identification of traffic events based on deep learning according to claim 5 is characterized in that: The method comprises the following steps: determining whether the currently reported traffic event is the same event as the most recently reported traffic event in the database; if so, performing duplicate traffic event reporting filtering; if not, performing normal reporting and updating the database reporting record; and Extract the appearance feature vector corresponding to the ROI image according to the event target of the traffic event confirmed by the secondary recognition model Retrieve and extract the appearance feature vector of the ROI image corresponding to the traffic event recently reported in the database Determine the appearance feature vector of the event target of the currently reported traffic event The event target appearance feature vector of the traffic incident recently reported in the database The minimum cosine distance Cos d ; If the minimum cosine distance Cos d If the distance is greater than the distance threshold, it is determined that the traffic event currently reported in the database record is the same as the traffic event most recently reported in the database, and duplicate traffic event reporting filtering is performed; otherwise, normal reporting is performed and the database reporting record is updated.

7. The method for secondary identification of traffic events based on deep learning according to claim 6, characterized in that: The minimum cosine distance Cos d Obtained by the following formula: in, represents the appearance feature vector of the event target of the currently reported traffic event, represents the appearance feature vector of the event target of the traffic event recently reported in the database, represents the dot product of the appearance feature vector of the event target of the currently reported traffic event and the appearance feature vector of the event target of the most recently reported traffic event in the database, and Represent the appearance feature vectors and The norm of .

8. The method for secondary identification of traffic events based on deep learning according to claim 3 is characterized in that: The process of classifying the types of traffic events and merging multiple levels of traffic events comprises the following steps: When the event target of the traffic incident is a person, the secondary recognition model classifies the event target of the traffic incident into social personnel, maintenance personnel or law enforcement personnel according to the functions of the personnel, and supplements the information of the event target; When the event target of the traffic incident is a vehicle, the secondary recognition model classifies the event target of the traffic incident into social vehicles, maintenance vehicles, law enforcement vehicles or hazardous chemicals vehicles according to the functions of the vehicles, and supplements the information of the event target; Merging multi-level traffic events according to the event target information of the traffic events by the secondary recognition model, wherein the multi-level traffic events include the same type of traffic events reported continuously in the spatiotemporal dimension and multiple types of traffic events reported continuously under a single pile number; The development process of the multi-level traffic event is displayed under the merged multi-level traffic event, and the progress snapshot photos and change process tracking videos of the multi-level traffic event are retained and recorded.

9. The method for secondary identification of traffic events based on deep learning according to claim 3, characterized in that: The determining whether the event type of the traffic event matches the relevant business rules, and if so, filtering the traffic event, comprises the following steps: Based on the event type of the traffic event determined by the secondary recognition model, determine all filtering rules corresponding to the event type; It is determined whether the event type of the traffic event belongs to any one of the filtering rules. If it matches, the traffic event is filtered. If it does not match, the traffic event is reported and confirmed for a second time.

10. A secondary traffic incident recognition system based on deep learning, characterized in that: include: The data collection module is provided with a detection and perception front end for real-time monitoring and collection of traffic information, and reporting the traffic incident when it is identified as a traffic incident; A data processing module, provided with a microprocessor and an embedded neural network processor, for receiving information of the traffic event and performing data processing; A data storage module, used to receive and store traffic event information to form an object storage database and a traffic event storage database; A data transmission module, used to realize communication transmission among the data acquisition module, the data processing module, the data storage module and the data center; The data processing module includes a traffic incident false alarm reminder unit, a repeated traffic incident reporting filtering unit, a multi-level traffic incident merging processing unit, and a traffic incident confirmation unit based on business logic; The traffic incident false alarm reminder unit is used to confirm the event target and event type of the traffic incident, and to confirm the event target using a secondary recognition model. If the event target of the traffic incident does not match the actual event target, a false alarm reminder is executed; The repeated traffic event reporting filtering unit is used to search the database for a traffic event to see if there is a report record with the same event type as the traffic event within a unit time. If not, normal reporting is performed and the database report record is updated. If so, repeated traffic event reporting filtering is performed. The multi-level traffic event merging processing unit is used to classify the event types of the traffic events and merge the multi-level traffic events; The traffic event confirmation unit based on business logic is used to determine whether the event type of the traffic event matches the relevant business rules. If it matches, the traffic event is filtered.

Images