Non-motor vehicle behavior monitoring method and system based on side cloud collaboration, and medium

Abstract

The invention discloses a non-motor vehicle behavior monitoring method and system based on side cloud collaboration and a medium, and belongs to the technical field of non-motor vehicle recognition. The method comprises an image acquisition preprocessing stage; a multi-frame image recognition processing stage: detecting position changes of the non-motor vehicle and the target person in continuous frames in real time, and recording pixel coordinate data in each frame to obtain a multi-frame recognition result; a multi-frame identification result transmission stage: transmitting the multi-frame identification result to a cloud computing platform in real time through a security protocol; a cloud behavior recognition and analysis stage: judging whether the non-motor vehicle and the target person have abnormal behaviors or not; and an intelligent alarm and response stage: when an abnormal behavior is detected, executing an alarm operation, sending alarm information to an administrator, and executing an abnormal intervention measure. A target detection algorithm and a multi-frame continuous behavior analysis method are adopted, real-time detection and accurate coordinate analysis are carried out on video streams, and high recognition accuracy can be kept in a complex corridor environment.

Description

Technical Field

[0001] The present invention relates to the field of non-motor vehicle identification technology, and in particular to a non-motor vehicle behavior monitoring method, system and medium based on edge-cloud collaboration. Background Art

[0002] Electric vehicles are favored by more and more consumers because of their small size, convenient use, long driving range and economical price. Especially in the case of increasingly congested urban traffic, electric vehicles have become the main means of transportation for many citizens. However, with the increasing number of electric vehicles, parking space resources have become extremely tight. Since urban building designs usually do not reserve enough parking space for electric vehicles, many residents have begun to choose to park electric vehicles in the corridor for the convenience of use, parking and charging. Although this practice solves the problem of temporary parking, it brings new safety hazards. There are many safety hazards in parking electric vehicles in residential corridors. First, electric vehicles occupy corridor space, which will directly block the escape route in fire or other emergency situations, seriously affecting the safe evacuation of personnel and the efficiency of fire rescue. Secondly, when electric vehicles are charged in the corridor, if the circuit is aging or the insulation layer is damaged, it is easy to cause a short circuit in the battery and cause a fire. Electric vehicles are usually flammable. Once a fire occurs, it will not only burn quickly and release a large amount of toxic smoke, but also cause chain combustion due to the narrow corridor, resulting in more serious casualties. Therefore, prohibiting the parking of electric vehicles in the corridor has become a management focus of many residential communities.

[0003] In order to prevent electric vehicles from being parked in corridors, local property management and government departments have taken a series of measures. The first is to raise residents' awareness of the hazards of parking electric vehicles in corridors through publicity and education, and to provide special parking spaces for electric vehicles. However, the actual effect is not ideal, and many residents still choose to park their electric vehicles in the corridors for convenience. Another measure is to arrange staff to regularly patrol the corridors, combining on-site inspections with remote video monitoring, but this method is costly to implement, and there is a risk of missed inspections in manual inspections. In addition, manual monitoring is not intelligent enough to achieve 24-hour real-time monitoring and timely response.

[0004] Most of the AI ​​recognition methods currently used in corridor monitoring rely on image feature recognition technology, and usually use a preset electric vehicle feature library to identify electric vehicles in the corridor. However, these systems often lack sufficient image preprocessing steps before recognition, such as denoising and contrast enhancement, which makes the recognition accuracy and robustness poor in low light or complex environments. Inaccurate recognition may occur, resulting in frequent false alarms and missed alarms. The current monitoring AI recognition method mainly focuses on the presence or absence of electric vehicles, and can only judge single-frame images. It cannot effectively distinguish between normal and abnormal behaviors, and fails to effectively identify the entry behavior of electric vehicles. This means that when the electric vehicle just enters the corridor, the system cannot issue an alarm in time, missing the best time to prevent the electric vehicle from entering the corridor. For corridor management, the most critical thing is not whether the electric vehicle is stationary, but whether an early warning can be issued at the moment the electric vehicle enters the corridor so as to intervene in time. Since the existing AI recognition method cannot distinguish the dynamic behavior of electric vehicles, frequent alarms for stationary electric vehicles are caused. Such repeated alarms not only easily make the property management become numb and ignore real alarms, but also make it impossible to connect the behavior to the work order system to assign orders to staff, reducing the credibility of the system and the effectiveness of the alarms.

[0005] Edge computing technology plays an important role in electric vehicle monitoring. It can process some data locally and reduce dependence on cloud computing. However, the computing power of edge devices is usually limited. When faced with high-resolution video data and complex image recognition tasks, they may not be able to respond in real time. This lack of computing power is particularly evident when the amount of data is large or the task is complex, affecting the real-time performance and response speed of the system.

[0006] In order to solve the problem of insufficient computing power of edge devices, many systems upload data to the cloud for processing. However, this method not only increases bandwidth costs, but also has the problem of data transmission delay. In addition, frequent data uploads may bring the risk of user privacy leakage, especially when the surveillance video involves public areas or residents' personal information. Therefore, how to balance the efficiency of data processing and user privacy protection in cloud computing is a difficult problem that needs to be solved urgently. Summary of the invention

[0007] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a non-motor vehicle behavior monitoring method, system and medium based on edge-cloud collaboration.

[0008] The objective of the present invention is achieved through the following technical solutions: The first aspect of the present invention provides: a non-motor vehicle behavior monitoring method based on edge-cloud collaboration, comprising the following steps: Image acquisition and preprocessing stage: real-time acquisition of the target area image, and preprocessing of the target area image by the edge computing device to obtain the first image; Multi-frame image recognition processing stage: Use a preset recognition algorithm to perform multi-frame recognition on the first image, detect the position changes of non-motor vehicles and target persons in continuous frames in real time, and record the pixel coordinate data in each frame to obtain multi-frame recognition results; Multi-frame recognition result transmission stage: The multi-frame recognition results are transmitted to the cloud computing platform in real time through a security protocol; Cloud behavior recognition and analysis stage: In the cloud computing platform, the integrated behavior recognition algorithm is used to deeply analyze the multi-frame recognition results to determine whether the non-motor vehicle and the target person have abnormal behavior; Intelligent alarm and response stage: When abnormal behavior is detected, the alarm operation is executed, the alarm information is sent to the administrator, and abnormal intervention measures are implemented.

[0009] Preferably, the preprocessing includes removing noise, adjusting contrast to a preset contrast, and adjusting illumination to a preset illumination intensity.

[0010] Preferably, the preset recognition algorithm includes a non-motor vehicle recognition algorithm and a target person recognition algorithm.

[0011] Preferably, the multi-frame recognition results are compressed and encrypted during the multi-frame recognition result transmission stage.

[0012] Preferably, the abnormal behavior is determined by the movement trajectories of the non-motor vehicle and the target person; the abnormal behavior includes non-motor vehicles entering the corridor, non-motor vehicles staying for a long time, and target persons parking illegally. Preferably, the warning information includes a description of the abnormal behavior, the time of occurrence, the specific location and relevant non-motor vehicle information.

[0013] Preferably, the cloud-based behavior recognition and analysis phase further includes the following steps: Calculate the center point and direction vector of the target person and non-motor vehicle in each frame; Calculate the direction angle between the target person and the non-motor vehicle. If the direction angle is less than the angle threshold, it is considered that the target person and the non-motor vehicle have the same moving direction. Detect whether the target person and the non-motor vehicle overlap or whether the minimum distance is less than the distance threshold. If the target person and the non-motor vehicle overlap or the minimum distance is less than the distance threshold, it is considered that the target person and the non-motor vehicle are together; When the target person and the non-motor vehicle move in the same direction and together within a continuous preset number of frames, it is determined that the target person and the non-motor vehicle enter the target area at the same time.

[0014] Preferably, each target person and non-motor vehicle is represented by a rectangle in each frame, and a rectangle is composed of four two-dimensional coordinate points [(x1, y1), (x2, y2), (x3, y3), (x4, y4)]; the center point of the rectangle is calculated by the following formula: Center(x, y) = ( (x1 + x2 + x3 + x4) / 4, (y1 + y2 + y3 + y4) / 4), wherein x1, y1 are the horizontal and vertical coordinates of the first vertex of the rectangle; x2, y2: the horizontal and vertical coordinates of the second vertex of the rectangle; x3, y3: the horizontal and vertical coordinates of the third vertex of the rectangle; x4, y4: the horizontal and vertical coordinates of the fourth vertex of the rectangle; Center(x, y): the horizontal and vertical coordinates of the center point of the rectangle; The direction vector of the rectangle is calculated by the following formula: DirectionVector(x, y) = (Center_current(x) - Center_previous(x), Center_current(y) -Center_previous(y)); Where Center_current(x, y) is the horizontal and vertical coordinates of the center point of the current frame; Center_previous(x, y) is the horizontal and vertical coordinates of the center point of the previous frame; DirectionVector(x, y) is the horizontal and vertical components of the direction vector; The direction angle between the target person and the non-motor vehicle is obtained by calculating the angle between the direction vectors of the target person and the non-motor vehicle; assuming that the direction vector of the target person is A = (Ax, Ay), and the direction vector of the non-motor vehicle is B = (Bx, By), the direction angle θ is calculated by the dot product: cos(θ) = (A • B) / (|A| |B|), where the calculation formula of the dot product A•B is: A•B=Ax*Bx+Ay*By; The magnitude of the direction vector is calculated by the following formula: |A|=√(Ax^2+Ay^2), |B|=√(Bx^2+By^2); The calculation formula of the direction angle θ is: θ = arccos((A•B) / (|A| |B|)); Wherein, Ax, Ay are the horizontal and vertical components of the direction vector of the target person; Bx, By are the horizontal and vertical components of the direction vector of the non-motor vehicle; |A| is the modulus of the direction vector of the target person; |B| is the modulus of the direction vector of the non-motor vehicle; θ is the direction angle between the two direction vectors of the target person and the non-motor vehicle; When it is detected that the distance between the center points of the two rectangles where the target person and the non-motor vehicle are located is less than the center point threshold, it is considered that there is overlap between the target person and the non-motor vehicle; if there is no overlap, the minimum distance between the vertex of the first rectangle closest to the second rectangle and the side segment of the second rectangle closest to the first rectangle is calculated. If the minimum distance is less than the distance threshold, it is considered that the target person and the non-motor vehicle are together.

[0015] The second aspect of the present invention provides: a non-motor vehicle behavior monitoring system based on edge-cloud collaboration, which is used to implement any of the above-mentioned non-motor vehicle behavior monitoring methods based on edge-cloud collaboration, comprising: An image acquisition preprocessing module is used to acquire a target area image in real time, and preprocess the target area image through an edge computing device to obtain a first image; A multi-frame image recognition processing module is used to perform multi-frame recognition on the first image using a preset recognition algorithm, detect the position changes of the non-motor vehicle and the target person in the continuous frames in real time, and record the pixel coordinate data in each frame to obtain the multi-frame recognition result; A multi-frame recognition result transmission module is used to transmit the multi-frame recognition results to the cloud computing platform in real time through a security protocol; The cloud-based behavior recognition and analysis module is used to conduct in-depth analysis of multi-frame recognition results through an integrated behavior recognition algorithm in the cloud computing platform to determine whether non-motor vehicles and target personnel have abnormal behaviors; The intelligent alarm and response module is used to perform alarm operations, send alarm information to administrators, and implement abnormal intervention measures when abnormal behavior is detected.

[0016] The third aspect of the present invention provides: a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are loaded and executed by a processor, any of the above-mentioned non-motor vehicle behavior monitoring methods based on edge-cloud collaboration is implemented.

[0017] The beneficial effects of the present invention are: 1) Adopting target detection algorithm and multi-frame continuous behavior analysis method, through real-time detection and precise coordinate analysis of video stream, it can maintain high recognition accuracy in complex corridor environment. Ensure that the alarm signal is triggered only when non-motor vehicles and personnel are confirmed to enter the corridor together, effectively distinguish normal behavior from abnormal behavior, reduce system interference, false alarms and missed alarms, and improve system reliability and recognition accuracy, user experience, and management efficiency. And through the edge-cloud collaborative architecture, the initial recognition of the video stream is performed on the edge side, and the coordinates and other character data are transmitted to the cloud for further behavior analysis, which significantly reduces the demand for network bandwidth and reduces transmission costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flow chart of the non-motor vehicle behavior monitoring method based on edge-cloud collaboration. DETAILED DESCRIPTION

[0019] 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.

[0020] See also Figure 1 The first aspect of the present invention provides: a non-motor vehicle behavior monitoring method based on edge-cloud collaboration, comprising the following steps: Image acquisition and preprocessing stage: real-time acquisition of the target area image, and preprocessing of the target area image by the edge computing device to obtain the first image; Multi-frame image recognition processing stage: Use a preset recognition algorithm to perform multi-frame recognition on the first image, detect the position changes of non-motor vehicles and target persons in continuous frames in real time, and record the pixel coordinate data in each frame to obtain multi-frame recognition results; Multi-frame recognition result transmission stage: The multi-frame recognition results are transmitted to the cloud computing platform in real time through a security protocol; Cloud behavior recognition and analysis stage: In the cloud computing platform, the integrated behavior recognition algorithm is used to deeply analyze the multi-frame recognition results to determine whether the non-motor vehicle and the target person have abnormal behavior; Intelligent alarm and response stage: When abnormal behavior is detected, the alarm operation is executed, the alarm information is sent to the administrator, and abnormal intervention measures are implemented.

[0021] In this embodiment, the camera installed in the corridor collects monitoring images in real time, and the collected images are preprocessed by the edge computing device. Image preprocessing includes noise removal, contrast enhancement, lighting adjustment, etc., to ensure that high-quality image input can be maintained under different lighting conditions, providing a good foundation for subsequent recognition steps. The pre-processed images are continuously recognized in multiple frames using the preset electric vehicle recognition algorithm and personnel recognition algorithm of the edge computing system. The edge computing device can detect the position changes of electric vehicles and personnel in continuous frames in real time, and record the pixel coordinate data in each frame. By analyzing multiple frames of images, the system can track the movement trajectory of electric vehicles and the activity status of personnel, and improve the accuracy and stability of recognition.

[0022] The results of multi-frame image recognition (including the target object and its pixel coordinates in each frame) are transmitted to the cloud computing platform in real time through a secure protocol. Efficient compression and encryption technology are used in the data transmission process to ensure data security while reducing transmission delays and improving the real-time response capability of the system. In the cloud computing platform, the data transmitted by the edge device is deeply analyzed through the integrated behavior recognition algorithm. The system uses machine learning models and combines the movement trajectories of electric vehicles and personnel to determine whether there are abnormal behaviors, such as electric vehicles entering the corridor, staying for a long time, or people parking illegally. This analysis can accurately locate the time and place of abnormal behavior, thereby improving the pertinence and effectiveness of the alarm. When the behavior recognition algorithm detects an abnormal situation, the system immediately triggers the alarm mechanism. The alarm information includes the description of the abnormal behavior, the time of occurrence, the specific location, and the relevant electric vehicle information. The alarm will be sent to property management personnel and relevant owners through multiple channels (such as property management system, mobile application or SMS) so that timely intervention measures can be taken. In addition, the system supports the setting of alarm priorities and the recording and analysis of historical alarms to help property managers conduct more effective security management.

[0023] In some embodiments, the preprocessing includes removing noise, adjusting contrast to a preset contrast, and adjusting illumination to a preset illumination intensity.

[0024] In some embodiments, the preset recognition algorithm includes a non-motor vehicle recognition algorithm and a target person recognition algorithm.

[0025] In some embodiments, the multi-frame recognition results are compressed and encrypted during the multi-frame recognition result transmission stage.

[0026] In some embodiments, the abnormal behavior is determined by the movement trajectories of the non-motor vehicle and the target person; the abnormal behavior includes a non-motor vehicle entering a corridor, a non-motor vehicle staying for a long time, and a target person parking illegally. In some embodiments, the warning information includes a description of the abnormal behavior, the time of occurrence, the specific location, and related non-motor vehicle information.

[0027] In some embodiments, the cloud-based behavior recognition and analysis phase further includes the following steps: Calculate the center point and direction vector of the target person and non-motor vehicle in each frame; Calculate the direction angle between the target person and the non-motor vehicle. If the direction angle is less than the angle threshold, it is considered that the target person and the non-motor vehicle have the same moving direction. Detect whether the target person and the non-motor vehicle overlap or whether the minimum distance is less than the distance threshold. If the target person and the non-motor vehicle overlap or the minimum distance is less than the distance threshold, it is considered that the target person and the non-motor vehicle are together; When the target person and the non-motor vehicle move in the same direction and together within a continuous preset number of frames, it is determined that the target person and the non-motor vehicle enter the target area at the same time.

[0028] In this embodiment, in order to comprehensively judge whether two entities (target person and non-motor vehicle) move together, it is necessary to set a distance threshold and an angle threshold. The distance threshold is used to judge whether the distance between the two entities is close enough. This value can be determined based on the entity size or relative movement speed in the actual scene. The angle threshold is used to judge whether the movement directions of the two entities are consistent, and is usually set to a value of 30 degrees or less. The specific threshold needs to be adjusted according to the application scenario.

[0029] In some embodiments, each target person and non-motor vehicle is represented by a rectangle in each frame, and a rectangle is composed of four two-dimensional coordinate points [(x1, y1), (x2, y2), (x3, y3), (x4, y4)]; the center point of the rectangle is calculated by the following formula: Center(x, y) = ( (x1 + x2 + x3 + x4) / 4, (y1 + y2 + y3 + y4) / 4), where x1, y1 are the horizontal and vertical coordinates of the first vertex of the rectangle; x2, y2: the horizontal and vertical coordinates of the second vertex of the rectangle; x3, y3: the horizontal and vertical coordinates of the third vertex of the rectangle; x4, y4: the horizontal and vertical coordinates of the fourth vertex of the rectangle; Center(x, y): the horizontal and vertical coordinates of the center point of the rectangle; The direction vector of the rectangle is calculated by the following formula: DirectionVector(x, y) = (Center_current(x) - Center_previous(x), Center_current(y) - Center_previous(y)); Where Center_current(x, y) is the horizontal and vertical coordinates of the center point of the current frame; Center_previous(x, y) is the horizontal and vertical coordinates of the center point of the previous frame; DirectionVector(x, y) is the horizontal and vertical components of the direction vector; The direction angle between the target person and the non-motor vehicle is obtained by calculating the angle between the direction vectors of the target person and the non-motor vehicle; assuming that the direction vector of the target person is A = (Ax, Ay), and the direction vector of the non-motor vehicle is B = (Bx, By), the direction angle θ is calculated by the dot product: cos(θ) = (A • B) / (|A| |B|), where the calculation formula of the dot product A•B is: A•B=Ax*Bx+Ay*By; The magnitude of the direction vector is calculated by the following formula: |A|=√(Ax^2+Ay^2), |B|=√(Bx^2+By^2); The calculation formula of the direction angle θ is: θ = arccos((A•B) / (|A| |B|)); Wherein, Ax, Ay are the horizontal and vertical components of the direction vector of the target person; Bx, By are the horizontal and vertical components of the direction vector of the non-motor vehicle; |A| is the modulus of the direction vector of the target person; |B| is the modulus of the direction vector of the non-motor vehicle; θ is the direction angle between the two direction vectors of the target person and the non-motor vehicle; When it is detected that the distance between the center points of the two rectangles where the target person and the non-motor vehicle are located is less than the center point threshold, it is considered that there is overlap between the target person and the non-motor vehicle; if there is no overlap, the minimum distance between the vertex of the first rectangle closest to the second rectangle and the side segment of the second rectangle closest to the first rectangle is calculated. If the minimum distance is less than the distance threshold, it is considered that the target person and the non-motor vehicle are together.

[0030] In this embodiment, the center point threshold is used to determine whether there is overlap between two entities. This value can be determined based on the size of the entity or the relative movement speed of the actual scene. Each entity (person or electric bicycle) is represented by a rectangle in each frame. In each frame, the coordinates of the entity are updated. It is necessary to analyze the changes in these coordinates between consecutive frames to determine the movement direction and relative position change of each entity. In order to represent the position of each rectangle (entity), we calculate its center point. The center point can be used to simplify subsequent distance calculation and direction analysis. Center point calculation: The center point (Center) represents the overall position of the rectangle. The calculation method is to add the x value and y value of the coordinates of the four corners of the rectangle respectively, and then divide it by 4. The direction vector (Direction Vector) represents the movement direction of the entity in two consecutive frames. The calculation method is to subtract the center point of the previous frame from the center point of the current frame to obtain a vector. When judging whether two entities enter the picture together, in addition to the direction similarity, their relative positions need to be checked. Overlap detection: Whether two rectangles overlap can be determined by detecting the distance between their center points. If the distance between the center points is less than a set threshold (such as the average width of the entity), it can be considered that the two rectangles overlap. Minimum distance calculation: If two rectangles do not overlap, the minimum distance between them also needs to be calculated.

[0031] The second aspect of the present invention provides: a non-motor vehicle behavior monitoring system based on edge-cloud collaboration, which is used to implement any of the above-mentioned non-motor vehicle behavior monitoring methods based on edge-cloud collaboration, comprising: An image acquisition preprocessing module is used to acquire a target area image in real time, and preprocess the target area image through an edge computing device to obtain a first image; A multi-frame image recognition processing module is used to perform multi-frame recognition on the first image using a preset recognition algorithm, detect the position changes of the non-motor vehicle and the target person in the continuous frames in real time, and record the pixel coordinate data in each frame to obtain the multi-frame recognition result; A multi-frame recognition result transmission module is used to transmit the multi-frame recognition results to the cloud computing platform in real time through a security protocol; The cloud-based behavior recognition and analysis module is used to conduct in-depth analysis of multi-frame recognition results through an integrated behavior recognition algorithm in the cloud computing platform to determine whether non-motor vehicles and target personnel have abnormal behaviors; The intelligent alarm and response module is used to perform alarm operations, send alarm information to administrators, and implement abnormal intervention measures when abnormal behavior is detected.

[0032] The third aspect of the present invention provides: a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are loaded and executed by a processor, any of the above-mentioned non-motor vehicle behavior monitoring methods based on edge-cloud collaboration is implemented.

[0033] 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 non-motor vehicle behavior monitoring method based on edge-cloud collaboration, characterized in that: The following steps are involved: Image acquisition and preprocessing stage: real-time acquisition of the target area image, and preprocessing of the target area image by the edge computing device to obtain the first image; Multi-frame image recognition processing stage: Use a preset recognition algorithm to perform multi-frame recognition on the first image, detect the position changes of non-motor vehicles and target persons in continuous frames in real time, and record the pixel coordinate data in each frame to obtain multi-frame recognition results; Multi-frame recognition result transmission stage: The multi-frame recognition results are transmitted to the cloud computing platform in real time through a security protocol; Cloud behavior recognition and analysis stage: In the cloud computing platform, the integrated behavior recognition algorithm is used to deeply analyze the multi-frame recognition results to determine whether the non-motor vehicle and the target person have abnormal behavior; Intelligent alarm and response stage: When abnormal behavior is detected, the alarm operation is executed, the alarm information is sent to the administrator, and abnormal intervention measures are implemented.

2. The non-motor vehicle behavior monitoring method based on edge-cloud collaboration according to claim 1 is characterized by: The preprocessing includes removing noise, adjusting contrast to a preset contrast, and adjusting illumination to a preset illumination intensity.

3. The non-motor vehicle behavior monitoring method based on edge-cloud collaboration according to claim 1 is characterized in that: The preset recognition algorithm includes a non-motor vehicle recognition algorithm and a target person recognition algorithm.

4. The non-motor vehicle behavior monitoring method based on edge-cloud collaboration according to claim 1 is characterized in that: The multi-frame recognition results are compressed and encrypted during the multi-frame recognition result transmission stage.

5. The non-motor vehicle behavior monitoring method based on edge-cloud collaboration according to claim 1 is characterized in that: The abnormal behavior is determined by the movement trajectories of the non-motor vehicle and the target person; the abnormal behavior includes non-motor vehicles entering the corridor, non-motor vehicles staying for a long time, and target persons parking illegally.

6. The non-motor vehicle behavior monitoring method based on edge-cloud collaboration according to claim 1 is characterized by: The warning information includes abnormal behavior description, occurrence time, specific location and related non-motor vehicle information.

7. The non-motor vehicle behavior monitoring method based on edge-cloud collaboration according to claim 1 is characterized by: The cloud-based behavior recognition and analysis phase also includes the following steps: Calculate the center point and direction vector of the target person and non-motor vehicle in each frame; Calculate the direction angle between the target person and the non-motor vehicle. If the direction angle is less than the angle threshold, it is considered that the target person and the non-motor vehicle have the same moving direction. Detect whether the target person and the non-motor vehicle overlap or whether the minimum distance is less than the distance threshold. If the target person and the non-motor vehicle overlap or the minimum distance is less than the distance threshold, it is considered that the target person and the non-motor vehicle are together; When the target person and the non-motor vehicle move in the same direction and together within a continuous preset number of frames, it is determined that the target person and the non-motor vehicle enter the target area at the same time.

8. The non-motor vehicle behavior monitoring method based on edge-cloud collaboration according to claim 7 is characterized by: Each target person and non-motor vehicle is represented by a rectangle in each frame. A rectangle consists of four two-dimensional coordinate points [(x1, y1), (x2, y2), (x3, y3), (x4, y4)]; the center point of the rectangle is calculated by the following formula: Center(x, y)= ( (x1 + x2 + x3 + x4) / 4,(y1 + y2 + y3 + y4) / 4), where x1, y1 are the horizontal and vertical coordinates of the first vertex of the rectangle; x2, y2: the horizontal and vertical coordinates of the second vertex of the rectangle; x3, y3: the horizontal and vertical coordinates of the third vertex of the rectangle; x4, y4: the horizontal and vertical coordinates of the fourth vertex of the rectangle; Center(x, y): the horizontal and vertical coordinates of the center point of the rectangle; The direction vector of the rectangle is calculated by the following formula: DirectionVector(x, y) = (Center_current(x)- Center_previous(x), Center_current(y)- Center_previous(y)); Where Center_current(x, y) is the horizontal and vertical coordinates of the center point of the current frame; Center_previous(x, y) is the horizontal and vertical coordinates of the center point of the previous frame; DirectionVector(x, y) is the horizontal and vertical components of the direction vector; The direction angle between the target person and the non-motor vehicle is obtained by calculating the angle between the direction vectors of the target person and the non-motor vehicle; assuming that the direction vector of the target person is A = (Ax, Ay), and the direction vector of the non-motor vehicle is B = (Bx, By), the direction angle θ is calculated by the dot product: cos(θ) = (A • B) / (|A| |B|), where the calculation formula of the dot product A•B is: A•B=Ax*Bx+Ay*By; The magnitude of the direction vector is calculated by the following formula: |A|=√(Ax^2+Ay^2), |B|=√(Bx^2+By^2); The calculation formula of the direction angle θ is: θ = arccos((A•B) / (|A| |B|)); Wherein, Ax, Ay are the horizontal and vertical components of the direction vector of the target person; Bx, By are the horizontal and vertical components of the direction vector of the non-motor vehicle; |A| is the modulus of the direction vector of the target person; |B| is the modulus of the direction vector of the non-motor vehicle; θ is the direction angle between the two direction vectors of the target person and the non-motor vehicle; When it is detected that the distance between the center points of the two rectangles where the target person and the non-motor vehicle are located is less than the center point threshold, it is considered that there is overlap between the target person and the non-motor vehicle; if there is no overlap, the minimum distance between the vertex of the first rectangle closest to the second rectangle and the side segment of the second rectangle closest to the first rectangle is calculated. If the minimum distance is less than the distance threshold, it is considered that the target person and the non-motor vehicle are together.

9. A non-motor vehicle behavior monitoring system based on edge-cloud collaboration, characterized by: The method for monitoring non-motor vehicle behavior based on edge-cloud collaboration according to any one of claims 1 to 8 comprises: An image acquisition preprocessing module is used to acquire a target area image in real time, and preprocess the target area image through an edge computing device to obtain a first image; A multi-frame image recognition processing module is used to perform multi-frame recognition on the first image using a preset recognition algorithm, detect the position changes of the non-motor vehicle and the target person in the continuous frames in real time, and record the pixel coordinate data in each frame to obtain the multi-frame recognition result; A multi-frame recognition result transmission module is used to transmit the multi-frame recognition results to the cloud computing platform in real time through a security protocol; The cloud-based behavior recognition and analysis module is used to conduct in-depth analysis of multi-frame recognition results through an integrated behavior recognition algorithm in the cloud computing platform to determine whether non-motor vehicles and target personnel have abnormal behaviors; The intelligent alarm and response module is used to perform alarm operations, send alarm information to administrators, and implement abnormal intervention measures when abnormal behavior is detected.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are loaded and executed by the processor, the non-motor vehicle behavior monitoring method based on edge-cloud collaboration as described in any one of claims 1-8 is implemented.

Images