Edge MEC event secondary analysis method and system and storage medium

Abstract

The invention discloses an edge MEC event secondary analysis method and system and a storage medium, and belongs to the technical field of intelligent traffic. The method comprises the following steps: S1, an equipment calibration stage; s2, a target tracking stage; s3, in an event detection stage, an event representation rule is defined, then whether a target event occurs in a target vehicle in the video sequence is judged according to the event representation rule, and if the target event occurs, recording is performed to obtain event representation; s4, an event secondary analysis stage: performing secondary analysis on the event representations, filtering out the event representations which do not pass the multiple authentication, and retaining the event representations which pass the multiple authentication to obtain final event representations; and S5, in an event transmission stage, uploading the final event representation to a service side for analysis and use. A deep learning model is used for carrying out secondary confirmation on an event target, secondary authentication is carried out on event key attributes such as position information and event types, event secondary analysis is automatically completed in an all-weather mode, manpower consumption is reduced, and the working efficiency of a service side is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to an edge MEC event secondary analysis method, system and storage medium. Background Art

[0002] The traditional event and accident disclosure logic mainly relies on manual inspections, which has low energy efficiency. Such a model is difficult to adapt to the trend of prosperous development of traffic and rapid growth of traffic flow. In the process of deep involvement of artificial intelligence in traffic scenes, event detection systems have been born. Specifically, the system uses expert systems to process the output of upstream algorithms (such as detection algorithms, tracking algorithms, etc.) and disclose predicted events to quickly respond to emergencies. Obviously, if the event detection system relies too much on the performance of upstream algorithms and lacks the ability to self-analyze and self-correct false alarms, there will be missed events and false alarms (such as wrong event target type, wrong lane, etc.). The event detection system in the existing technology has a small system perception range, and the interaction relationship between the vehicle and the computer and the panoramic environment are difficult to obtain. Event analysis can only be performed on the vehicle that uploads data. Summary of the invention

[0003] The purpose of the present invention is to overcome the deficiencies of the prior art and to provide an edge MEC event secondary analysis method, system and storage medium.

[0004] The objective of the present invention is achieved through the following technical solutions: The first aspect of the present invention provides: a secondary analysis method of edge MEC events, comprising the following steps: S1: Equipment calibration stage, use the calibration method to calibrate a single device or multiple devices to eliminate distortion; S2: Target tracking stage, using the target detection algorithm to perform the target detection task in the video sequence to obtain the vehicle target detection result, and then continuously track the target vehicle in the vehicle target detection result in the video sequence; S3: Event detection stage, define event representation rules, and then judge whether the target event occurs in the target vehicle in the video sequence according to the event representation rules. If the target event occurs, record it and obtain the event representation; S4: Secondary analysis of events: secondary analysis of event representations, filtering out event representations that have not passed multiple authentications, retaining event representations that have passed multiple authentications, and obtaining the final event representations; S5: In the event transmission stage, the final event representation is uploaded to the business side for analysis and use.

[0005] Preferably, before executing S1: device calibration phase, it is also necessary to set up a camera and a millimeter wave radar.

[0006] Preferably, the target detection task includes a target positioning subtask and a target classification subtask; the target positioning subtask is used to confirm the position and size of a vehicle in a certain frame of a video sequence; and the target classification subtask is used to confirm the category of the vehicle.

[0007] Preferably, event representation rules are defined using object-based representation and / or scene-based representation.

[0008] Preferably, the event representation includes the target event, occurrence time, occurrence location, and event behavior description. Preferably, the S4: event secondary analysis stage further includes the following steps: In the event target authentication stage, a pre-trained deep learning model is used to determine whether the target event in the event representation is a preset event. If not, it is filtered out. If it is, it is retained to obtain the first event representation. In the position attribute authentication stage, the first event representation is input into the position relocation model, and the position attribute of the target event represented by the first event is corrected through the environmental data; the vehicle driving direction, all the coil data of the point, and the coordinates of the detection frame are used as environmental data, and the key reference area of ​​the detection frame is determined according to the vehicle driving direction. The position is relocated according to the IOU results of the key reference area and all the coil data to obtain the second event representation; In the event category authentication stage, the second event representation is subjected to event behavior correction, and then authenticated using a deep learning model and environmental data. If the authentication is successful, the final event representation is obtained.

[0009] Preferably, the lane is detected and lane attributes are annotated to obtain coil data through a lane line automatic detection model.

[0010] Preferably, the calibration method is Zhang Zhengyou calibration method; the target detection algorithm is Faster R-CNN or YOLO or SSD algorithm.

[0011] The second aspect of the present invention provides: an edge MEC event secondary analysis system, used to implement any of the above-mentioned edge MEC event secondary analysis methods, including: The device calibration module is used to calibrate a single device or multiple devices using a calibration method to eliminate distortion; The target tracking module is used to perform a target detection task in a video sequence using a target detection algorithm to obtain a vehicle target detection result, and then continuously track the target vehicle in the vehicle target detection result in the video sequence; The event detection module is used to define event representation rules, and then determine whether the target event occurs in the target vehicle in the video sequence according to the event representation rules. If the target event occurs, it is recorded to obtain the event representation; The event secondary analysis module is used to perform secondary analysis on the event representation, filter out the event representation that has not passed the multiple authentications, and retain the event representation that has passed the multiple authentications to obtain the final event representation; The event transmission module is used to upload the final event representation to the business side for analysis and use.

[0012] The third aspect of the present invention provides: a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are loaded and executed by a processor, any of the above-mentioned edge MEC event secondary analysis methods is implemented.

[0013] The beneficial effects of the present invention are: 1) In road-side perception scenarios (cameras, radars), an automated secondary authentication process is performed on the event representation output of event detection. A deep learning model is used to reconfirm the event target and to reauthenticate key event attributes such as location information and event type. Secondary event analysis is completed automatically around the clock, reducing manpower consumption and improving work efficiency on the business side. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 Flow chart of the secondary analysis method for edge MEC events; Figure 2 This is a schematic diagram of vehicle detection results; Figure 3 This is a multi-authentication flow chart. DETAILED DESCRIPTION

[0015] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0016] The present invention uses road-side perception to observe relevant road sections in real time. Compared with vehicle-side perception, the road-side perceives and collects upstream events in a larger range and conducts secondary analysis. Combined with the deep model, an automated event secondary analysis process is implemented to ensure erroneous analysis, including non-traffic event target filtering, lane relocation, event type correction, etc.

[0017] See also Figure 1-Figure 3 The first aspect of the present invention provides: a secondary analysis method for edge MEC events, comprising the following steps: S1: Equipment calibration stage, use the calibration method to calibrate a single device or multiple devices to eliminate distortion; S2: Target tracking stage, using the target detection algorithm to perform the target detection task in the video sequence to obtain the vehicle target detection result, and then continuously track the target vehicle in the vehicle target detection result in the video sequence; S3: Event detection stage, define event representation rules, and then judge whether the target event occurs in the target vehicle in the video sequence according to the event representation rules. If the target event occurs, record it and obtain the event representation; S4: Secondary analysis of events: secondary analysis of event representations, filtering out event representations that have not passed multiple authentications, retaining event representations that have passed multiple authentications, and obtaining the final event representations; S5: In the event transmission stage, the final event representation is uploaded to the business side for analysis and use.

[0018] In this embodiment, a secondary analysis platform for traffic scene events is designed using deep learning models and road-side perception data, which observes relevant road sections in real time, perceives environmental data, collects upstream events and conducts secondary analysis, and implements an automated all-weather secondary analysis process for events, including non-traffic event target filtering, lane relocation, event type correction, etc. The system input is the output of upstream detection, tracking and other algorithms, and then uses a deep learning model to automatically authenticate event targets and related event types. If the non-event target or event-related attribute is wrong, the event is filtered or the relevant attribute is modified before reporting the event. In order to filter fewer event targets by mistake while ensuring the filtering performance of non-event targets, the model FDA-Net is trained in the field of traffic using event targets to distinguish event targets from non-event targets. The samples recognized by the model as non-event targets are automatically filtered, and the remaining targets are sent to the event-related attribute confirmation module, which includes lane attribute confirmation and event type confirmation, which are responsible for correcting lane attributes and confirming event types respectively. This system is based on road-side perception, and while efficiently utilizing road-side data, it improves the accuracy of event detection and the work efficiency of inspectors, and reduces the cost of manual screening.

[0019] In S2, target tracking refers to continuously tracking a specific target in a video sequence and determining the position of the target in each frame. Unlike target detection, target tracking requires continuous tracking of the target in images of different frames. In S3, event detection refers to automatically detecting events that people are interested in or concerned about in the video, such as traffic violations by pedestrians, goals in sports games, etc., aiming to use visual data and machine learning technology to automatically perceive and analyze events that occur in the real world. In order to complete event detection, it is first necessary to define event representation, that is, how to encode and represent events. Commonly used methods include object-based representation and scene-based representation. Event detection refers to how to detect target events from the video and define them according to the pre-defined event representation. This step receives inputs of target detection and tracking and environmental perception factors, and finally determines whether an event occurs, and if so, it is revealed.

[0020] In some embodiments, it is necessary to set up a camera and a millimeter-wave radar before executing S1: the device calibration phase.

[0021] In some embodiments, the target detection task includes a target positioning subtask and a target classification subtask; the target positioning subtask is used to confirm the position and size of a vehicle in a certain frame of a video sequence; and the target classification subtask is used to confirm the category of the vehicle.

[0022] In some embodiments, event representation rules are defined using object-based representation and / or scene-based representation.

[0023] In some embodiments, the event representation includes the target event, occurrence time, occurrence location, and event behavior description. In some embodiments, the S4: event secondary analysis phase further includes the following steps: In the event target authentication stage, a pre-trained deep learning model is used to determine whether the target event in the event representation is a preset event. If not, it is filtered out. If it is, it is retained to obtain the first event representation. In the position attribute authentication stage, the first event representation is input into the position relocation model, and the position attribute of the target event represented by the first event is corrected through the environmental data; the vehicle driving direction, all the coil data of the point, and the coordinates of the detection frame are used as environmental data, and the key reference area of ​​the detection frame is determined according to the vehicle driving direction. The position is relocated according to the IOU results of the key reference area and all the coil data to obtain the second event representation; In the event category authentication stage, the second event representation is subjected to event behavior correction, and then authenticated using a deep learning model and environmental data. If the authentication is successful, the final event representation is obtained.

[0024] In this embodiment, the lane is detected and the lane attributes (such as lane type, etc.) are marked to form a coil through the lane line automatic detection model, and then the lane is located through the relationship between the detection frame and the coil. Event behavior correction includes: after the position is relocated, if the lane attributes change, the event type is re-determined according to the change.

[0025] In some embodiments, the lane is detected and lane attributes are annotated to obtain coil data through a lane line automatic detection model.

[0026] In some embodiments, the calibration method is Zhang Zhengyou calibration method; the target detection algorithm is Faster R-CNN or YOLO or SSD algorithm.

[0027] The second aspect of the present invention provides: an edge MEC event secondary analysis system, used to implement any of the above-mentioned edge MEC event secondary analysis methods, including: The device calibration module is used to calibrate a single device or multiple devices using a calibration method to eliminate distortion; The target tracking module is used to perform a target detection task in a video sequence using a target detection algorithm to obtain a vehicle target detection result, and then continuously track the target vehicle in the vehicle target detection result in the video sequence; The event detection module is used to define event representation rules, and then determine whether the target event occurs in the target vehicle in the video sequence according to the event representation rules. If the target event occurs, it is recorded to obtain the event representation; The event secondary analysis module is used to perform secondary analysis on the event representation, filter out the event representation that has not passed the multiple authentications, and retain the event representation that has passed the multiple authentications to obtain the final event representation; The event transmission module is used to upload the final event representation to the business side for analysis and use.

[0028] The third aspect of the present invention provides: a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are loaded and executed by a processor, any of the above-mentioned edge MEC event secondary analysis methods is implemented.

[0029] The above is only a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, and should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be modified within the scope of the concept described herein through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art shall not deviate from the spirit and scope of the present invention, and shall be within the scope of protection of the claims attached to the present invention.

Claims

1. A secondary analysis method for edge MEC events, characterized by: The following steps are involved: S1: Equipment calibration stage, use the calibration method to calibrate a single device or multiple devices to eliminate distortion; S2: Target tracking stage, using the target detection algorithm to perform the target detection task in the video sequence to obtain the vehicle target detection result, and then continuously track the target vehicle in the vehicle target detection result in the video sequence; S3: Event detection stage, define event representation rules, and then judge whether the target event occurs in the target vehicle in the video sequence according to the event representation rules. If the target event occurs, record it and obtain the event representation; S4: Secondary analysis of events: secondary analysis of event representations, filtering out event representations that have not passed multiple authentications, retaining event representations that have passed multiple authentications, and obtaining the final event representations; S5: In the event transmission stage, the final event representation is uploaded to the business side for analysis and use.

2. The edge MEC event secondary analysis method according to claim 1, characterized in that: Before executing S1: Equipment calibration phase, it is necessary to set up the camera and millimeter wave radar.

3. The edge MEC event secondary analysis method according to claim 1, characterized in that: The target detection task includes a target positioning subtask and a target classification subtask; The target positioning subtask is used to confirm the position and size of a vehicle in a certain frame of a video sequence; the target classification subtask is used to confirm the category of the vehicle.

4. The edge MEC event secondary analysis method according to claim 1, characterized in that: Event representation rules are defined using object-based representation and / or scene-based representation.

5. The edge MEC event secondary analysis method according to claim 1, characterized in that: The event representation includes the target event, occurrence time, occurrence location, and event behavior description.

6. The edge MEC event secondary analysis method according to claim 5, characterized in that: The S4: event secondary analysis phase also includes the following steps: In the event target authentication phase, a pre-trained deep learning model is used to determine whether the target event in the event representation is a preset event. If not, it is filtered out. If it is, it is retained to obtain the first event representation. In the position attribute authentication stage, the first event representation is input into the position relocation model, and the position attribute of the target event represented by the first event is corrected through the environmental data; the vehicle driving direction, all the coil data of the point, and the coordinates of the detection frame are used as environmental data, and the key reference area of ​​the detection frame is determined according to the vehicle driving direction. The position is relocated according to the IOU results of the key reference area and all the coil data to obtain the second event representation; In the event category authentication stage, the second event representation is subjected to event behavior correction, and then authenticated using a deep learning model and environmental data. If the authentication is successful, the final event representation is obtained.

7. The edge MEC event secondary analysis method according to claim 6, characterized in that: The lane line automatic detection model is used to detect lanes and annotate lane attributes to obtain coil data.

8. The edge MEC event secondary analysis method according to any one of claims 1 to 7, characterized in that: The calibration method is Zhang Zhengyou calibration method; the target detection algorithm is Faster R-CNN or YOLO or SSD algorithm.

9. An edge MEC event secondary analysis system, characterized by: The method for implementing the secondary analysis of edge MEC events according to any one of claims 1 to 8 comprises: The device calibration module is used to calibrate a single device or multiple devices using a calibration method to eliminate distortion; The target tracking module is used to perform a target detection task in a video sequence using a target detection algorithm to obtain a vehicle target detection result, and then continuously track the target vehicle in the vehicle target detection result in the video sequence; The event detection module is used to define event representation rules, and then determine whether the target event occurs in the target vehicle in the video sequence according to the event representation rules. If the target event occurs, it is recorded to obtain the event representation; The event secondary analysis module is used to perform secondary analysis on the event representation, filter out the event representation that has not passed the multiple authentications, and retain the event representation that has passed the multiple authentications to obtain the final event representation; The event transmission module is used to upload the final event representation to the business side for analysis and use.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are loaded and executed by the processor, the edge MEC event secondary analysis method as described in any one of claims 1-8 is implemented.

Images