An edge MEC event secondary analysis method and system and a storage medium

Abstract

The application 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, a device calibration stage; S2, a target tracking stage; S3, an event detection stage, defining an event representation rule, then judging whether a target vehicle in a video sequence has a target event according to the event representation rule, if the target vehicle has the target event, recording to obtain an event representation; S4, an event secondary analysis stage, performing secondary analysis on the event representation, filtering out event representations that do not pass multiple authentications, and retaining event representations that pass multiple authentications to obtain final event representations; and S5, an event exposure stage, uploading the final event representations to a business side for analysis and use. The deep learning model is used for secondary confirmation of an event target, secondary authentication of event key attributes such as position information and an event type, automatic all-weather completion of event secondary analysis, reduction of human consumption, and improvement of the working efficiency of the business side.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and in particular to a method, system, and storage medium for secondary analysis of edge MEC events. Background Technology

[0002] Traditional event detection methods rely heavily on manual patrols, resulting in low efficiency and an inability to adapt to the booming transportation industry and rapidly increasing traffic flow. As artificial intelligence (AI) becomes more deeply integrated into traffic scenarios, event detection systems have emerged. Specifically, these systems use expert systems to process the output of upstream algorithms (such as detection and tracking algorithms) and generate predicted events for rapid response to emergencies. However, if event detection systems overly depend on the performance of upstream algorithms and lack the ability to self-analyze and correct false alarms, they will experience missed or false alarms (such as incorrect event target type or lane error). Current event detection systems have a limited sensing range, making it difficult to capture the interaction between vehicles and the system, and the overall environment; they can only analyze events from the vehicle's uploaded data. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, system and storage medium for secondary analysis of edge MEC events.

[0004] The objective of this invention is achieved through the following technical solution: The first aspect of this invention provides: a method for secondary analysis of edge MEC events, comprising the following steps:

[0005] S1: Equipment calibration stage, using calibration methods to calibrate one or more pieces of equipment to eliminate distortion;

[0006] S2: In the target tracking stage, a target detection algorithm is used to perform a target detection task in the video sequence to obtain vehicle target detection results, and then the target vehicles in the vehicle target detection results are continuously tracked in the video sequence.

[0007] S3: Event detection phase. Define event representation rules, and then determine whether the target vehicle in the video sequence has experienced a target event based on the event representation rules. If a target event has occurred, record it to obtain the event representation.

[0008] S4: Secondary event analysis stage. The event representation is analyzed a second time to filter out event representations that have not passed multi-factor authentication and retain event representations that have passed multi-factor authentication to obtain the final event representation;

[0009] S5: Event disclosure phase, the final event representation is uploaded to the business side for parsing and use.

[0010] Preferably, a camera and millimeter-wave radar need to be installed before performing S1: the equipment calibration stage.

[0011] Preferably, the target detection task includes a target localization subtask and a target classification subtask; the target localization subtask is used to determine the position and size of the vehicle in a certain frame of the video sequence; the target classification subtask is used to determine the category of the vehicle.

[0012] Preferably, event representation rules are defined using object-based and / or scene-based representations.

[0013] Preferably, the event representation includes the target event, the time of occurrence, the location of occurrence, and a description of the event behavior.

[0014] Preferably, the S4: secondary event analysis stage further includes the following steps:

[0015] 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 it is not, it is filtered out; if it is, it is retained to obtain the first event representation.

[0016] In the location attribute authentication stage, the first event representation is input into the location relocation model, and the location attribute of the target event in the first event representation is corrected by environmental data; the vehicle driving direction, all coil data of the point, and the coordinates of the detection box are used as environmental data, the key reference area of ​​the detection box is determined according to the vehicle driving direction, and the location relocation is performed based on the IOU results of the key reference area and all coil data to obtain the second event representation.

[0017] In the event category authentication phase, the second event representation undergoes event behavior correction, and then authentication is performed using a deep learning model and environmental data. If the authentication is successful, the final event representation is obtained.

[0018] Preferably, the lanes are detected and lane attributes are labeled using an automatic lane line detection model to obtain coil data.

[0019] Preferably, the calibration method is the Zhang Zhengyou calibration method; the target detection algorithm is Faster R-CNN, YOLO, or SSD algorithm.

[0020] A second aspect of the present invention provides: a secondary analysis system for edge MEC events, used to implement any of the above-mentioned secondary analysis methods for edge MEC events, comprising:

[0021] The equipment calibration module is used to calibrate one or more devices using calibration methods to eliminate distortion.

[0022] The target tracking module is used to perform target detection tasks in the video sequence using target detection algorithms to obtain vehicle target detection results, and then continuously track the target vehicles in the vehicle target detection results in the video sequence.

[0023] The event detection module is used to define event representation rules, and then determine whether a target event has occurred in the video sequence based on the event representation rules. If a target event has occurred, it is recorded to obtain the event representation.

[0024] The event secondary analysis module is used to perform secondary analysis on the event representation, filtering out event representations that have not passed multi-factor authentication, and retaining event representations that have passed multi-factor authentication to obtain the final event representation;

[0025] The event exposure module is used to upload the final event representation to the business side for parsing and use.

[0026] A third aspect of the present invention provides: a computer-readable storage medium storing computer-executable instructions, wherein when the computer-executable instructions are loaded and executed by a processor, any of the above-described edge MEC event secondary analysis methods are implemented.

[0027] The beneficial effects of this invention are:

[0028] 1) In roadside perception scenarios (cameras, radar), an automated secondary authentication process is performed on the event representation output of event detection. A deep learning model is used to confirm the event target and perform secondary authentication on key event attributes such as location information and event type. The secondary analysis of events is completed automatically around the clock, reducing manpower consumption and improving the work efficiency of the business side. Attached Figure Description

[0029] Figure 1 Flowchart of the secondary analysis method for edge MEC events;

[0030] Figure 2 This is a diagram illustrating the vehicle inspection results.

[0031] Figure 3 This is a flowchart of the multi-factor authentication process. Detailed Implementation

[0032] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] This invention uses roadside perception to observe relevant road sections in real time. Compared with vehicle-side perception, roadside perception can perceive a wider range and collect upstream events for secondary analysis. Combined with deep models, it implements an automated event secondary analysis process, ensuring that erroneous analysis includes filtering of non-traffic event targets, lane relocation, and event type correction.

[0034] See Figures 1-3 The first aspect of this invention provides: a method for secondary analysis of edge MEC events, comprising the following steps:

[0035] S1: Equipment calibration stage, using calibration methods to calibrate one or more pieces of equipment to eliminate distortion;

[0036] S2: In the target tracking stage, a target detection algorithm is used to perform a target detection task in the video sequence to obtain vehicle target detection results, and then the target vehicles in the vehicle target detection results are continuously tracked in the video sequence.

[0037] S3: Event detection phase. Define event representation rules, and then determine whether the target vehicle in the video sequence has experienced a target event based on the event representation rules. If a target event has occurred, record it to obtain the event representation.

[0038] S4: Secondary event analysis stage. The event representation is analyzed a second time to filter out event representations that have not passed multi-factor authentication and retain event representations that have passed multi-factor authentication to obtain the final event representation;

[0039] S5: Event disclosure phase, the final event representation is uploaded to the business side for parsing and use.

[0040] In this embodiment, a traffic scene event secondary analysis platform is designed using deep learning models and roadside perception data. This platform monitors relevant road sections in real time, senses environmental data, collects upstream events, and performs secondary analysis, implementing an automated, 24 / 7 event secondary analysis process, including non-traffic event target filtering, lane relocation, and event type correction. The system takes as input the output of upstream detection and tracking algorithms, then uses a deep learning model to automatically authenticate event targets and related event types. If a non-event target or its related attributes are incorrect, the event is filtered or its attributes are modified before being reported. To ensure filtering performance for non-event targets while minimizing false positives, a model FDA-Net, trained on event targets in the traffic domain, is used to distinguish between event and non-event targets. Samples identified as non-event targets are automatically filtered, while other targets are sent to an event-related attribute confirmation module, including lane attribute confirmation and event type confirmation, which are responsible for correcting lane attributes and confirming event types, respectively. This system, based on roadside perception, efficiently utilizes roadside data, improves event detection accuracy and patrol personnel efficiency, and reduces manual screening costs.

[0041] In S2, target tracking refers to continuously tracking a specific target in a video sequence and determining the target's position in each frame. Unlike target detection, target tracking requires continuous tracking of the target across different frames. In S3, event detection refers to automatically detecting events of interest or concern in a video, such as pedestrian traffic violations or goals in sports matches. It aims to automatically perceive and analyze real-world events using visual data and machine learning techniques. To perform event detection, it is first necessary to define an event representation—that is, how to encode and represent events. Common methods include object-based representation and scene-based representation. Event detection refers to how to detect target events from a video and define them according to a predefined event representation. This step receives input from target detection and tracking, as well as environmental perception factors, and ultimately determines whether an event has occurred; if so, it is revealed.

[0042] In some embodiments, a camera and millimeter-wave radar need to be installed before performing S1: the device calibration phase.

[0043] In some embodiments, the target detection task includes a target localization subtask and a target classification subtask; the target localization subtask is used to determine the position and size of a vehicle in a certain frame of a video sequence; the target classification subtask is used to determine the category of the vehicle.

[0044] In some embodiments, event representation rules are defined using object-based representations and / or scene-based representations.

[0045] In some embodiments, the event representation includes the target event, the time of occurrence, the location of occurrence, and a description of the event behavior.

[0046] In some embodiments, S4: the secondary event analysis phase, further includes the following steps:

[0047] 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 it is not, it is filtered out; if it is, it is retained to obtain the first event representation.

[0048] In the location attribute authentication stage, the first event representation is input into the location relocation model, and the location attribute of the target event in the first event representation is corrected by environmental data; the vehicle driving direction, all coil data of the point, and the coordinates of the detection box are used as environmental data, the key reference area of ​​the detection box is determined according to the vehicle driving direction, and the location relocation is performed based on the IOU results of the key reference area and all coil data to obtain the second event representation.

[0049] In the event category authentication phase, the second event representation undergoes event behavior correction, and then authentication is performed using a deep learning model and environmental data. If the authentication is successful, the final event representation is obtained.

[0050] In this embodiment, an automatic lane detection model is used to detect lanes and label lane attributes (such as lane type) to form loops. Then, the lane is located based on the relationship between the detection frame and the loops. Event behavior correction includes: after relocation, if lane attributes change, the event type is re-determined based on this change.

[0051] In some embodiments, a lane line automatic detection model is used to detect lanes and label lane attributes to obtain coil data.

[0052] In some embodiments, the calibration method is the Zhang Zhengyou calibration method; the target detection algorithm is Faster R-CNN, YOLO, or SSD algorithm.

[0053] A second aspect of the present invention provides: a secondary analysis system for edge MEC events, used to implement any of the above-mentioned secondary analysis methods for edge MEC events, comprising:

[0054] The equipment calibration module is used to calibrate one or more devices using calibration methods to eliminate distortion.

[0055] The target tracking module is used to perform target detection tasks in the video sequence using target detection algorithms to obtain vehicle target detection results, and then continuously track the target vehicles in the vehicle target detection results in the video sequence.

[0056] The event detection module is used to define event representation rules, and then determine whether a target event has occurred in the video sequence based on the event representation rules. If a target event has occurred, it is recorded to obtain the event representation.

[0057] The event secondary analysis module is used to perform secondary analysis on the event representation, filtering out event representations that have not passed multi-factor authentication, and retaining event representations that have passed multi-factor authentication to obtain the final event representation;

[0058] The event exposure module is used to upload the final event representation to the business side for parsing and use.

[0059] A third aspect of the present invention provides: a computer-readable storage medium storing computer-executable instructions, wherein when the computer-executable instructions are loaded and executed by a processor, any of the above-described edge MEC event secondary analysis methods are implemented.

[0060] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for secondary analysis of edge MEC events, characterized in that: Includes the following steps: S1: Equipment calibration stage, using calibration methods to calibrate one or more pieces of equipment to eliminate distortion; S2: In the target tracking stage, a target detection algorithm is used to perform a target detection task in the video sequence to obtain vehicle target detection results, and then the target vehicles in the vehicle target detection results are continuously tracked in the video sequence. S3: Event detection phase. Define event representation rules, and then determine whether the target vehicle in the video sequence has experienced a target event based on the event representation rules. If a target event has occurred, record it to obtain the event representation. S4: Secondary event analysis stage. The event representation is analyzed a second time to filter out event representations that have not passed multi-factor authentication and retain event representations that have passed multi-factor authentication to obtain the final event representation; S5: Event disclosure phase, the final event representation is uploaded to the business side for parsing and use; The event representation includes the target event, the time of occurrence, the location of occurrence, and a description of the event behavior; The S4 stage, the secondary event analysis stage, 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 it is not, it is filtered out; if it is, it is retained to obtain the first event representation. In the location attribute authentication stage, the first event representation is input into the location relocation model, and the location attribute of the target event represented by the first event is corrected through environmental data. Using the vehicle driving direction, all coil data of the point, and the coordinates of the detection box as environmental data, the key reference area of ​​the detection box is determined according to the vehicle driving direction. The location relocation is performed based on the IOU results of the key reference area and all coil data to obtain the second event representation. Among them, the lane line automatic detection model is used to detect the lane and label the lane attributes to obtain the coil data. In the event category authentication phase, the second event representation undergoes event behavior correction, and then authentication is performed using a deep learning model and environmental data. If the authentication is successful, the final event representation is obtained.

2. The edge MEC event secondary analysis method according to claim 1, characterized in that: Before performing the S1: equipment calibration phase, cameras and millimeter-wave radar need to be installed.

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

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

5. The edge MEC event secondary analysis method according to any one of claims 1-4, characterized in that: The calibration method is the Zhang Zhengyou calibration method; the target detection algorithm is Faster R-CNN, YOLO, or SSD algorithm.

6. A secondary analysis system for edge MEC events, characterized in that: For implementing the edge MEC event secondary analysis method as described in any one of claims 1-5, comprising: The equipment calibration module is used to calibrate one or more devices using calibration methods to eliminate distortion. The target tracking module is used to perform target detection tasks in the video sequence using target detection algorithms to obtain vehicle target detection results, and then continuously track the target vehicles in the vehicle target detection results in the video sequence. The event detection module is used to define event representation rules, and then determine whether a target event has occurred in the video sequence based on the event representation rules. If a target event has occurred, it is recorded to obtain the event representation. The event secondary analysis module is used to perform secondary analysis on the event representation, filtering out event representations that have not passed multi-factor authentication, and retaining event representations that have passed multi-factor authentication to obtain the final event representation; The event exposure module is used to upload the final event representation to the business side for parsing and use.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which, when loaded and executed by a processor, implement the edge MEC event secondary analysis method as described in any one of claims 1-5.

Images