Charging parking space parking standard identification system and method based on target detection and coordinate geometry priori knowledge

Abstract

The invention discloses a charging parking space parking specification identification system and method based on target detection and coordinate geometry priori knowledge, and the system has good real-time performance and coverage continuity through the grouping management and timing frame extraction of cameras. Through image distortion correction, perspective correction and brightness enhancement, the image quality is improved, and the interference of a complex environment on an identification result is reduced; the target detection model outputs the positions and categories of the vehicles and different types of license plates at the same time, the blue-licensed vehicles and the new energy vehicles can be accurately distinguished, and the problem that the blue-licensed vehicles occupy charging parking spaces and cannot be automatically recognized is fundamentally solved; the point position judgment based on the ray method and the line pressing judgment based on the minimum distance make the judgment of the space relation between the vehicle and the parking space more stable and reliable, and misjudgment caused by camera angle change, detection frame drifting or illumination change is effectively avoided; and in combination with a geometric rule of polygonal coordinates of the parking space, the system has relatively high scene adaptability.

Description

Technical Field

[0001] This application relates to the fields of computer vision and artificial intelligence, and in particular to a charging parking space parking regulation recognition system and method based on target detection and coordinate geometry prior knowledge. Background Technology

[0002] With the accelerated implementation of global energy transition strategies, the number of electric vehicles is growing rapidly, and the scale of charging infrastructure construction is constantly expanding. Charging parking spaces have become a crucial link supporting the healthy development of the electric vehicle industry. The management efficiency of charging parking spaces not only affects the utilization efficiency of charging facilities but also directly impacts the charging experience for electric vehicle users and the overall operational order of parking lots. In typical scenarios such as shopping malls, office buildings, and residential communities, how to intelligently sense and regulate the usage of charging parking spaces is gradually becoming an important research direction in the fields of smart parking and smart transportation.

[0003] However, under current technological conditions, the management of charging parking spaces still relies primarily on traditional methods such as manual inspections and fixed signage. On the one hand, the illegal occupation of charging spaces by non-charging vehicles (such as gasoline vehicles) is widespread. Relying on security personnel for regular patrols is not only labor-intensive but also difficult to detect and stop short-term parking occupancy in a timely manner. On the other hand, even electric vehicles themselves often park improperly, such as parking over the lines or with the vehicle body significantly offset from the center of the parking space, which can easily lead to conflicts between adjacent parking spaces or affect the normal access of charging facilities. In recent years, although some image recognition solutions based on video surveillance have been attempted for parking behavior analysis, they are mostly limited to simple target detection or static area occupancy judgment, failing to effectively integrate multi-dimensional information such as vehicle type (such as license plate color and energy attribute) and parking space geometric boundaries. At the same time, some solutions rely on high-performance cloud servers and complex deep learning models, which face problems such as high network bandwidth pressure, high computing costs, and difficulty in model deployment in visual network scenarios, making it difficult to stably promote and apply them on a large scale and in multiple scenarios.

[0004] Therefore, in the identification and management of parking regulations for charging spaces, it is difficult to identify non-charging vehicles illegally occupying parking spaces in a timely and accurate manner, it is difficult to make precise judgments on the parking posture of charging vehicles in combination with the geometric boundaries of the parking space, and it is difficult to deploy intelligent analysis systems at low cost and with wide coverage under the condition of limited cloud network resources. These are problems that urgently need to be solved. Summary of the Invention

[0005] This application provides a system and method for identifying parking regulations in charging parking spaces based on target detection and prior knowledge of coordinate geometry. It aims to solve the problems in the identification and management of parking regulations in charging parking spaces, such as the difficulty in timely and accurate identification of illegal occupation by non-charging vehicles, the difficulty in accurately determining the parking posture of charging vehicles in combination with the geometric boundaries of the parking space, and the difficulty in deploying intelligent analysis systems at low cost and with wide coverage under the condition of limited cloud network resources.

[0006] In a first aspect, a charging parking space parking standard recognition system based on target detection and coordinate geometry prior knowledge is provided. The system includes: a video inspection strategy module, a target detection model module, and an irregular parking recognition module. The video inspection strategy module is used to manage the access of cameras corresponding to charging spaces, establish the association between cameras and inspection tasks, and perform timed frame extraction on the video stream according to a preset inspection cycle to obtain frame images. The target detection model module is used to perform target detection on the frame-sampling image to obtain target detection results including vehicle targets and license plate targets. The target detection results include target category labels and bounding box coordinates. The irregular parking identification module is used to: analyze the spatial relationship between the vehicle center point or license plate center point and the parking space area based on the target detection results and the coordinate geometry prior knowledge base, so as to identify the vehicle parking behavior of crossing the line and / or the illegal parking behavior of blue-plate vehicles, and generate an event record with event information. The coordinate geometry prior knowledge base is used to store the polygon coordinate information of the parking space area and the coordinate geometry rules used for judging irregular parking behavior.

[0007] Optionally, in the above scheme, the video inspection strategy module includes a camera adaptation unit and a task configuration unit. The camera adaptation unit is used to obtain the resolution, frame rate, installation position and orientation of the camera and divide the camera into different inspection task groups according to preset rules. The task configuration unit is used to set the inspection time interval for different inspection task groups.

[0008] Optionally, in the above scheme, the video inspection strategy module further includes an image processing unit, which is used to perform distortion correction, perspective correction, brightness enhancement or contrast enhancement processing on the extracted frame images, and generate image tags for identification and association.

[0009] Optionally, in the above scheme, the target detection model module adopts a lightweight multi-scale target detection model and is configured to output the category labels and corresponding bounding box coordinates of vehicle targets, blue license plate targets, and green license plate targets.

[0010] Optionally, in the above scheme, the non-standard parking recognition module is used to determine whether the center point of the vehicle or the center point of the license plate is located inside the polygonal area of ​​the parking space using the ray method, and calculates the minimum distance between the vehicle center point and each side of the parking space when the center point of the vehicle is located inside the area to determine whether it is crossing the line.

[0011] Secondly, a method for recognizing parking regulations in charging spaces based on target detection and prior knowledge of coordinate geometry, the method comprising: Obtain the camera configuration and perform timed frame extraction on the camera's video stream according to the inspection strategy to obtain the extracted frame images; Perform image correction and image enhancement processing on the extracted frames; The processed frame-segmented image is input into the target detection model to obtain target detection results that include vehicle targets and license plate targets. Read the polygon coordinate information and coordinate geometry determination rules of the corresponding parking space; Based on the target detection results and prior knowledge of coordinate geometry, the spatial relationship between the vehicle center point or license plate center point and the parking space area is determined in order to identify vehicle parking behavior that crosses the line and / or illegal parking behavior of blue-plate vehicles. Event logs are generated when irregular parking behavior is detected.

[0012] Optionally, in the above scheme, obtaining the camera configuration and performing timed frame extraction on the video stream of the camera according to the inspection strategy to obtain the extracted frame image includes: dividing the camera into a camera group for inspecting vehicles that cross the line and parking, and a camera group for inspecting illegal parking of blue-plate vehicles, according to the camera's resolution, frame rate, installation angle and orientation, and configuring the inspection cycle for each group.

[0013] Optionally, the image processing in the image correction and image enhancement processing performed on the frame-dropped image includes distortion correction, perspective correction, brightness enhancement, and contrast enhancement.

[0014] Optionally, in the above scheme, determining the spatial relationship between the vehicle center point or license plate center point and the parking space area includes: calculating the vehicle center point or license plate center point based on the target bounding box coordinates, and using the ray method to determine whether the center point is located inside the polygon of the parking space area.

[0015] Optionally, the method further includes: when the center point of the vehicle is located inside the parking space area, calculating the distance from the center point of the vehicle to each side of the parking space and selecting the minimum distance, comparing it with a preset distance threshold to determine whether the vehicle meets the line crossing condition.

[0016] Compared with the prior art, this application has at least the following beneficial effects: Based on further analysis and research of existing technical problems, this application recognizes that in the identification and management of parking regulations in charging parking spaces, it is difficult to identify non-charging vehicles illegally occupying parking spaces in a timely and accurate manner, it is difficult to make precise judgments on the parking posture of charging vehicles in combination with the geometric boundaries of the parking space, and it is difficult to deploy intelligent analysis systems at low cost and with wide coverage under the condition of limited cloud network resources. By combining video inspection strategies, image preprocessing, target detection and coordinate geometric prior knowledge, this application achieves accurate identification of the parking status of vehicles in the charging parking space area, which can effectively solve the problems of low identification accuracy, high false judgment rate, strong dependence on the environment and inability to distinguish vehicle types in existing technologies. By grouping and managing cameras and performing timed frame extraction, the system achieves excellent real-time performance and coverage continuity. Image distortion correction, perspective correction, and brightness enhancement improve image quality and reduce interference from complex environments on recognition results. The target detection model simultaneously outputs the position and category of vehicles and different types of license plates, accurately distinguishing between blue-plate vehicles and new energy vehicles, fundamentally solving the problem of automatically identifying blue-plate vehicles occupying charging parking spaces. Point-based determination based on ray casting and line-crossing determination based on minimum distance make the spatial relationship determination between vehicles and parking spaces more stable and reliable, effectively avoiding misjudgments caused by changes in camera angle, detection frame drift, or lighting changes. Combining the geometric rules of parking space polygon coordinates gives the system strong scene adaptability. In summary, this system can accurately identify parking over the line and illegal parking of blue-plate vehicles, significantly improving parking management capabilities and solving the problems of inaccurate recognition, insufficient stability, and low processing efficiency mentioned in the background technology. Attached Figure Description

[0017] Figure 1 A schematic diagram of the framework of a charging parking space parking rule recognition system based on target detection and coordinate geometry prior knowledge provided in one embodiment of this application; Figure 2 This is a schematic diagram illustrating target detection category labeling as provided in one embodiment of this application; Figure 3 A schematic diagram representing the coordinate geometric prior knowledge of a vehicle stopping on the line according to an embodiment of this application; Figure 4 A schematic diagram representing the coordinate geometric prior knowledge of illegally parked blue-plate vehicles provided in one embodiment of this application; Figure 5 This is a schematic diagram of a vehicle parking on the line according to one embodiment of this application; Figure 6 This is a schematic diagram of a blue-plate vehicle illegally parked, provided as an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] In the description of this application: unless otherwise stated, "a plurality of" means two or more. The terms "first," "second," "third," etc., in this application are intended to distinguish the objects referred to and do not have any special meaning in terms of technical connotation (e.g., they should not be construed as an emphasis on importance or order). Expressions such as "including," "comprising," and "having" also mean "not limited to" (certain units, components, materials, steps, etc.).

[0020] With the accelerated implementation of global energy transition strategies, the electric vehicle industry has entered a critical period of large-scale application. As a new type of infrastructure supporting industry development, the management efficiency of charging parking spaces directly impacts user experience and resource utilization. Currently, the industry faces a dual management dilemma: the unauthorized occupation by non-charging vehicles and the improper parking of electric vehicles. The disorderly occupation by non-charging vehicles leads to significantly lower-than-expected utilization of charging facilities, while the out-of-bounds movement of charging vehicles causes secondary problems such as equipment connection failures and parking space conflicts.

[0021] Faced with these thorny issues, traditional management methods are limited by a passive regulatory model, relying on manual inspections and static signage, making it difficult to achieve real-time detection and dynamic intervention of violations. Existing technical solutions are mostly based on low-dimensional perception systems, and simple image recognition methods cannot effectively analyze the correlation between vehicle energy attributes and spatial posture. Furthermore, fixed threshold detection mechanisms are difficult to adapt to the differentiated layout requirements of different parking lots.

[0022] In recent years, with the rapid development of artificial intelligence technology, significant breakthroughs have been achieved in the field of target detection, among which the YOLOv5 model has attracted much attention. This invention, in its exploration of standardized parking management for charging parking spaces, introduces the YOLOv5 model. This model can quickly and accurately identify various targets in extracted video frames from parking lots, accurately outputting the coordinate information of the targets. Furthermore, it deeply integrates the detected target coordinate information with the coordinate geometry knowledge related to the parking space. For example, in the extracted image frame, a digital model is pre-constructed based on coordinate geometry parameters such as the actual size of the parking space, the common dimensions of cars, and the relative positional relationship between the parking space and surrounding passageways. When YOLOv5 detects a gasoline vehicle parked in a charging parking space, the system can quickly compare the license plate coordinates of the vehicle with the coordinate range of the charging parking space. Once it determines that the vehicle is in an area where it should not be parked, an alarm mechanism is immediately triggered, thus providing strong technical support for achieving intelligent and precise standardized parking management, and is expected to completely change the current chaotic situation of charging parking space management.

[0023] The fusion of deep learning technology YOLOv5 with coordinate geometric information offers several significant advantages. Firstly, in terms of detection accuracy, traditional detection methods, relying solely on simple image recognition, struggle to accurately distinguish between vehicle types and precise parking space matching. However, by incorporating coordinate geometric information, combining the vehicle target information identified by YOLOv5 with the standard dimensions of the parking space, the system accurately determines whether a vehicle is parked legally, significantly reducing the false positive rate. This allows for accurate identification of parking status for various vehicle types, from small electric cars to large SUVs. Secondly, regarding real-time performance, YOLOv5's inherent ability to process images quickly, coupled with the integration of coordinate geometric information eliminating the need for complex secondary calculations, enables the system to instantly determine vehicle information and parking space compliance. Compared to traditional manual checks, this allows for immediate alerts upon the occurrence of violations, greatly improving management efficiency. Third, it is highly adaptable. Faced with the diverse layouts of different parking lots, such as irregularly shaped parking spaces and complex passageways, YOLOv5 can quickly adapt and flexibly adjust its detection strategies with the help of pre-built physical models. This ensures that it can stably and accurately identify parking regulations in various scenarios, providing a solid guarantee for the orderly management of charging parking spaces.

[0024] This embodiment fully considers the practical problems in visual network scenarios, such as high cloud network costs and the difficulty in implementing artificial intelligence models due to the complexity of the scenarios. It integrates object detection models with interpretable coordinate geometric prior knowledge through programming, achieving intelligent recognition of irregular parking behavior with a low-computing-cost and interpretable technical solution. The successful pilot implementation of this solution in a charging station scenario demonstrates that this system can be replicated and promoted to scenarios such as large shopping malls, office buildings, and residential communities.

[0025] In this embodiment, as Figure 1 As shown, a charging parking space parking standard recognition system based on target detection and coordinate geometry prior knowledge is provided. The system includes: a video inspection strategy module, a target detection model module, and an irregular parking recognition module. The video inspection strategy module is used to manage the access of cameras corresponding to charging spaces, establish the association between cameras and inspection tasks, and perform timed frame extraction on the video stream according to a preset inspection cycle to obtain frame images. The target detection model module is used to perform target detection on the frame-sampling image to obtain target detection results including vehicle targets and license plate targets. The target detection results include target category labels and bounding box coordinates. The irregular parking identification module is used to: analyze the spatial relationship between the vehicle center point or license plate center point and the parking space area based on the target detection results and the coordinate geometry prior knowledge base, so as to identify the vehicle parking behavior of crossing the line and / or the illegal parking behavior of blue-plate vehicles, and generate an event record with event information. The coordinate geometry prior knowledge base is used to store the polygon coordinate information of the parking space area and the coordinate geometry rules used for judging irregular parking behavior.

[0026] This embodiment provides a charging parking space parking standard recognition system based on target detection and coordinate geometry prior knowledge. The system includes a video inspection strategy module, a target detection model module, and an irregular parking recognition module. The modules are linked through software call relationships and data interfaces to realize automatic inspection and behavior recognition of vehicle parking status in the charging parking space area.

[0027] In its implementation, the video inspection strategy module first manages the access of multiple cameras. This module obtains basic parameters for each camera, such as resolution, frame rate, installation height, tilt angle, focal length, and the number of parking spaces covered, through the camera adaptation unit. Based on the analysis of the charging parking space monitoring scenario, it assigns the cameras to either the vehicle crossing the line detection task group or the blue-plate vehicle illegal parking detection task group. Subsequently, the system sets independent frame extraction periods for different task groups through the task configuration unit. For example, a shorter frame extraction period is set for vehicle crossing the line detection, while a moderate frame extraction period is set for blue-plate vehicle illegal parking, balancing real-time performance and computational cost. When the period arrives, the system automatically retrieves the real-time video stream from the corresponding camera and performs frame extraction, obtaining a extracted frame image containing a timestamp and camera number.

[0028] Before object detection, the image processing unit preprocesses the extracted frames, including but not limited to distortion correction, perspective correction, brightness enhancement, and contrast enhancement. Distortion correction compensates for barrel or pincushion distortion caused by the camera lens; perspective correction restores the tilted image to a projection closer to a top-down angle, ensuring the accuracy of the parking space geometry; brightness enhancement and contrast enhancement ensure that the object detection model can reliably extract vehicle edge contours and license plate information even in low-light or shadow environments. The processed images are assigned a unique identifier for recording and retrospective analysis after identifying irregular parking behavior.

[0029] The target detection model module employs a lightweight multi-scale target detection network, such as a YOLO-based model. Through a combination of a backbone network, feature pyramid, and detection head, it achieves simultaneous detection of vehicle targets, blue license plate targets, and green license plate targets. The structured data output by the model includes the category label for each target (e.g., car, blue_plate, green_plate), bounding box coordinates (x1, y1, x2, y2), and confidence score. Since charging parking spaces are special-purpose parking spaces, the detection results of blue license plate vehicles will serve as an important input for subsequent identification of illegally parked blue license plate vehicles.

[0030] The coordinate geometry prior knowledge base stores the coordinates of the polygon vertices of the parking space area, such as the four vertices A, B, C, and D of the parking space, organized in a clockwise or counterclockwise manner. Simultaneously, the knowledge base also stores the ray-mapping rules for determining if a point lies within the polygon, as well as the minimum distance threshold Z required for a vehicle to park on the line. The parking space boundary lines can be further converted into standard straight line equations, facilitating subsequent calculations of the distance from the center point to the boundary.

[0031] The irregular parking detection module, based on target detection results and geometric data in a prior knowledge base, first calculates the vehicle's center point using the target bounding box coordinates. It then uses a ray casting method to determine if the vehicle's center point is inside the parking space polygon. When the vehicle is inside the parking space, the system further calculates the perpendicular distances from the center point to each side of the parking space and compares the minimum distance with a preset threshold Z to determine if the vehicle is parked beyond the line. For license plate targets, the system determines whether a blue-plate vehicle occupies a charging parking space based on whether the license plate's center point enters the parking space area. The identified irregular parking behaviors are ultimately recorded by the event generation unit. These records include the behavior type, event time, camera number, and corresponding image identifier, facilitating platform presentation and subsequent processing.

[0032] By adopting the above technical solution, this embodiment can significantly overcome many problems existing in the prior art: First, by introducing a multi-camera inspection strategy and image preprocessing mechanism, the problem of unstable recognition caused by changes in camera angle, large field of view distortion, or insufficient lighting in traditional monitoring methods is solved, making image features such as vehicle bounding boxes, license plates, and parking space edges clearer, providing stable input for subsequent algorithms.

[0033] Secondly, by using a target detection model to achieve automatic vehicle and license plate recognition, the shortcomings of traditional methods based on pure geometric thresholds in distinguishing different vehicle types and license plate attributes are avoided. This enables the system to automatically recognize the behavior of blue-plate vehicles entering charging parking spaces, improving the level of intelligence in the management of dedicated electric vehicle parking spaces.

[0034] Third, this embodiment constructs a polygonal coordinate system for parking spaces and introduces the ray method to determine whether a point is located inside the parking space. This makes vehicle position determination no longer dependent on a single threshold or fixed area division, solving the problem of traditional solutions being sensitive to the camera installation angle and orientation, and enhancing the system's adaptability to different site layouts.

[0035] Fourth, this embodiment identifies the behavior of crossing the line based on the minimum distance calculation method from the center point to the boundary of the parking space, avoiding the problems of "boundary jitter" and "accumulation of segmentation result error" common in the background technology, and effectively reducing the risk of misjudgment caused by the offset of the detection box.

[0036] Fifth, by generating event records and associating them with frame-by-frame images, this embodiment can achieve automatic prompting and recording management of vehicles parking on the line and blue-plate vehicles occupying parking spaces, shortening the long cycle and discontinuous coverage of traditional manual inspections, and enabling the parking management system to respond to on-site situations more efficiently and accurately.

[0037] This embodiment can not only accurately identify various irregular parking behaviors such as parking over the line and illegal parking of vehicles with blue license plates, but also adapt to different installation conditions and environmental changes, thereby providing stable and reliable technical support for urban parking facility management and solving key problems such as low recognition accuracy, insufficient stability and slow dynamic response in the prior art.

[0038] In this embodiment, the video inspection strategy module includes a camera adaptation unit and a task configuration unit. The camera adaptation unit is used to obtain the resolution, frame rate, installation position and orientation of the camera and divide the camera into different inspection task groups according to preset rules. The task configuration unit is used to set the inspection time interval for different inspection task groups.

[0039] In this embodiment, the video inspection strategy module further includes an image processing unit, which is used to perform distortion correction, perspective correction, brightness enhancement or contrast enhancement processing on the extracted frame images, and generate image tags for identification and association.

[0040] In this embodiment, the target detection model module adopts a lightweight multi-scale target detection model and is configured to output the category labels and corresponding bounding box coordinates of vehicle targets, blue license plate targets, and green license plate targets.

[0041] In this embodiment, the irregular parking recognition module is used to determine whether the center point of the vehicle or the center point of the license plate is located inside the polygonal area of ​​the parking space using the ray method, and calculates the minimum distance between the vehicle center point and each side of the parking space when the center point of the vehicle is located inside the area to determine whether it is over the line.

[0042] Current charging space applications face several challenges: First, the illegal occupation of charging spaces by non-electric vehicles is difficult to monitor and remove in real time through manual inspections, leading to charging difficulties for electric vehicles and wasting charging resources. A precise and rapid method to identify and warn against illegal occupation by non-electric vehicles is urgently needed. Second, the improper parking of electric vehicles makes it difficult to accurately analyze their position and posture using existing technologies to ensure proper parking. Third, the significant differences between different parking lots mean that existing solutions lack universality and cannot guarantee the rational use of charging spaces. To address these issues, this paper proposes a charging space parking compliance identification method based on object detection and coordinate geometry prior knowledge to solve the aforementioned technical challenges and identify improper parking behaviors such as vehicles parking over the lines and vehicles with blue license plates illegally parking. This method integrates the lightweight object detection algorithm YOLOv5 with an interpretable algorithm based on coordinate geometry prior knowledge for improper parking identification. Details are as follows: The target detection model is responsible for detecting all target objects representing irregular parking behavior. This method can detect all target objects required for irregular parking behavior identification using only a lightweight multi-scale target detection model, which greatly reduces the computing power cost required by the model and improves the effectiveness of inspection.

[0043] By incorporating prior knowledge of coordinate geometry into the detection of target objects (such as vehicles and license plates) and the analysis of video frame extraction, an AI function for identifying irregular parking, which is technically extremely challenging for visual models, was achieved through an interpretable and verifiable algorithm design. This interpretable method not only boasts high accuracy but also exhibits better robustness by not requiring massive amounts of data to learn model parameters.

[0044] The purpose of this embodiment is to solve two major technical problems in the visual network scenario: "high cloud network cost" and "difficulty in implementing artificial intelligence model technology due to the complexity of the scenario". It aims to achieve intelligent recognition of parking regulations in charging sites with weak basic conditions (poor network, hardware, and monitoring environment) and complex scenarios.

[0045] On the one hand, it helps promote the application of this patented technology in places with charging spaces, such as shopping mall parking lots and community parking lots; on the other hand, through the low-cost and high-accuracy solution proposed by this patent, it can, to a certain extent, replace manpower to complete the video inspection of charging spaces, which not only helps the supervision of charging spaces but also reduces the management costs of enterprises.

[0046] This embodiment provides a charging parking space parking standard recognition system based on target detection and coordinate geometry prior knowledge, which consists of four modules: non-standard parking behavior, video inspection strategy, target detection model, and non-standard parking recognition algorithm based on coordinate geometry prior knowledge. Figure 1 A framework diagram of a charging parking space parking regulation recognition system based on object detection and coordinate geometry prior knowledge is presented. In the diagram, non-compliant parking behaviors include vehicles parking over the lines and illegal parking of vehicles with blue license plates. For such inspection tasks, the video inspection strategy module is responsible for implementing functions such as multi-camera access, timed inspection task mechanism, and camera-task association. Next, after the inspection strategy is issued, the object detection model is requested to obtain object category and location information, where the detection capability is trained using the detection category data designed in this scheme. Then, the non-compliant parking recognition algorithm is implemented using the coordinate geometry prior knowledge proposed in this scheme, and after analysis, the results are asynchronously returned to the requester and a warning is issued via a pop-up window on the monitoring screen. Further, the design of each module of the system is as follows: Irregular Parking Behavior. This module is a demand transformation matrix, abstracting the demand for irregular parking in charging spaces into achievable AI problems. Irregular parking behaviors include: vehicles parking over the lines and illegal parking of vehicles with blue license plates. Specifically: vehicles parking over the lines can be transformed into a problem of AI detecting and analyzing vehicles and license plates in sampled images. The AI ​​needs to be able to identify whether a vehicle exceeds the boundary line of the parking space, which involves object detection technology in computer vision. By recognizing vehicle information and using prior geometric knowledge of parking space coordinates, it determines whether a vehicle has crossed the line. Illegal parking of vehicles with blue license plates can be transformed into a problem of AI recognizing the color of the license plate. Using object detection technology in computer vision, the AI ​​system analyzes the color information of the license plate and combines it with prior geometric knowledge of parking space coordinates to determine whether a vehicle with a blue license plate is parked in a charging space.

[0047] Video Inspection Strategy. This module is responsible for converting the monitoring cameras in charging locations into the inspection strategy of this solution. This includes connecting the cameras to the video network platform and completing the standardized collection of basic camera information through the platform, covering dimensions such as camera resolution, frame rate, installation location, orientation, and network transmission protocol, laying the foundation for adaptation analysis. Subsequently, based on the task requirements of "vehicles parking on the line" and "illegally parked vehicles with blue license plates," cameras are adapted in a graded manner: for "vehicles parking on the line," which requires accurate capture of vehicle location information and the relative position of the parking space boundary, cameras directly above the parking space with a resolution of ≥1080P and capable of clearly capturing the parking space line and the edge of the vehicle are prioritized for adaptation; for "illegally parked vehicles with blue license plates," which requires clear capture of the license plate color and the parking space, cameras to the side / directly in front of the parking space and capable of clearly capturing the license plate color and the parking space are prioritized for adaptation. Finally, through software development, the association mechanism between the cameras and the corresponding inspection tasks is mapped to the platform parameter configuration management module to achieve efficient resource utilization and convenient maintenance in the future.

[0048] Based on the inspection time requirements, the timed inspection task mechanism of the cameras is mapped to the platform parameter configuration management module through software development. Relevant personnel can set the timed task for automatic inspection of each irregular parking behavior through the platform parameter management module. For example, the timed inspection time for the parking over the line inspection task corresponding to camera number 1 is set to every 30 minutes, and the timed inspection time for the blue-plate vehicle illegal parking inspection task corresponding to camera number 2 is set to every 15 minutes. 3) Because the backend requests AI model services in a concurrent manner, and some AI inspection tasks need to define parameters such as video frame extraction frequency, analysis window length, target association rules, and conclusion judgment thresholds based on inspection requirements; then, video frames are collected in a targeted manner and preprocessed with image correction, feature enhancement, etc., and each group of data is labeled with a unique identifier; then, target detection is performed on continuous frame images to extract static features such as vehicles and license plates and make a preliminary judgment based on prior knowledge of parking space coordinates; then, the judgment threshold is used to confirm whether the vehicle is illegally parked; finally, the conclusion is output and returned to the backend. Therefore, there is a real problem of long AI inference time. Therefore, an asynchronous mechanism is used to realize the message connection between the backend and the AI ​​model to ensure that the service does not time out and the data is transmitted in a consistent manner.

[0049] For object detection, this method employs the lightweight object detection model YOLOv5 and integrates the detection results with interpretable coordinate geometric prior knowledge, effectively identifying irregularly parked vehicles. The English and Chinese labels for the object detection categories are shown in the table below, with examples of labeling provided. Figure 2 As shown.

[0050]

[0051] An algorithm for identifying irregular parking based on coordinate geometry prior knowledge. This embodiment treats the conditions upon which irregular parking behavior interacts with the environment as interpretable coordinate geometry prior knowledge with low uncertainty. This prior knowledge is used to determine the target detection results, effectively identifying irregular parking behavior in complex scenes and significantly reducing GPU resource requirements. For details regarding the coordinate geometry prior knowledge described in this method based on the irregular parking behavior requirement matrix, please refer to [link to relevant documentation]. Figure 3 and Figure 4 As shown. Specifically, the prior knowledge existing in the scene image includes: 1) the center point P of the vehicle target, 2) the physical coordinate information of the parking space, 3) the coordinate range of the vehicle crossing the line, and 4) the center point P of the license plate. Figure 1 This method combines the detection results of the detection model in module (3) with the coordinate geometric prior knowledge in module (4) to develop an algorithm for identifying irregular parking. This algorithm is interpretable and verifiable. It should be noted that the coordinate geometric prior knowledge for judging irregular parking is as follows: Figure 3 As shown, let the four vertices of the parking space be arranged in order as A. B C D The center point of the vehicle is P. To determine if a vehicle's center point is within a parking space using the ray casting method, first, draw a ray horizontally to the right from the vehicle's center point P. Count the number of intersections between the ray and the parking space boundary. For each parking space edge (e.g., AB, BC, CD, DA), determine if the ray intersects it. If there is an intersection and the x-coordinate of the intersection point is greater than [value missing], then [the ray is considered to be within the parking space boundary]. If the number of intersections is odd, the vehicle's center point is inside the parking space. If the number of intersections is even, the vehicle's center point P is outside the parking space. If the vehicle's center point P is inside the parking space and the distance Δx from any parking space edge is less than the threshold Z, it is considered to be over the line. If the center point P of a blue-plate vehicle's license plate is within the parking space, regardless of its distance from the edge line, it is considered to be illegally parked.

[0052] The pseudo-algorithm for vehicle lane crossing and parking alarm recognition is designed as follows: Inputs: camera and task association mechanism, temporal information, spatial information, target detection model, and annotation rules relevant to this task; process: 1. Start a scheduled task to inspect the list of event cameras (charging parking area) and extract video addresses; 2: The extracted frames are fed into the object detection model to obtain the detection result det_reco; 3: If the vehicle (car) exists in det_reco, the analysis will begin, and the preconditions are met; 4: If a vehicle is detected crossing the line and the duration does not meet the threshold T, repeat steps 2-4. 5: If a vehicle is detected crossing the line and the duration exceeds the threshold T, stop the vehicle crossing the line. 6: Then append ["event_behavior": vehicle stopped on the line,"event_time":xx,"event_camera":xxx,"event_image":xxxx] to response_result; 7: Return response_result # Initialize to empty; Output: response_result #json, exception information (event timestamp, any image from the event frame, camera ID, etc.).

[0053] The pseudo-algorithm for identifying illegally parked blue-plate vehicles is designed as follows: Inputs: Camera and task association mechanism, temporal information, spatial information, object detection model, and annotation rules relevant to this task. process: 1. Start a scheduled task to inspect the list of event cameras (charging parking area) and extract video addresses; 2: The extracted frames are fed into the object detection model to obtain the detection result det_reco; 3: If (blue_plate) exists in det_reco and # starts the analysis, the prerequisite is met; 4: If a blue license plate (blue_plate) is detected in a parking space and the duration does not meet the threshold T, repeat steps 2-4; 5: If a blue license plate (blue_plate) is detected in a parking space and the duration exceeds the threshold T, the blue license plate vehicle is illegally parked. 6: Then append ["event_behavior":Blue-plate vehicle illegally parked,"event_time":xx,"event_camera":xxx,"event_image":xxxx] to response_result; 7: Return response_result # Initialize to empty; Output: response_result #json, exception information (event timestamp, any image from the event frame, camera ID, etc.).

[0054] This embodiment addresses the technical challenges posed by the complex environment of charging parking spaces, poor camera installation locations and lighting conditions, and the need for continuous video analysis to identify improper parking. It proposes a method for identifying proper parking in charging parking spaces based on object detection and coordinate geometry prior knowledge. This method solves the aforementioned technical challenges and enables the identification of improper parking behaviors such as vehicles parking over the lines and vehicles with blue license plates illegally parked. The method combines the lightweight object detection algorithm YOLOv5 with an interpretable algorithm based on coordinate geometry prior knowledge for identifying improper parking. This interpretable method offers high accuracy and, more importantly, does not require massive amounts of data to learn model parameters, thus exhibiting better robustness.

[0055] The purpose of this embodiment is to address the technical challenges posed by high cloud network costs and complex scenarios in video network environments, which hinder the implementation of artificial intelligence models. It aims to intelligently identify the proper parking of charging parking spaces under conditions of poor network environment, inadequate hardware, poor monitoring environment, and complex scenarios. This facilitates the replication and promotion of this method in locations with charging parking spaces, such as shopping malls, office buildings, and residential communities. Furthermore, the low-cost and high-accuracy solution proposed in this embodiment can, to some extent, replace manual labor in video inspections of improperly parked charging parking spaces, thus aiding in the supervision of charging parking services and reducing enterprise management costs. The improper parking identification system for charging parking spaces designed based on this patent has been applied in multiple real-world scenarios. Figure 5 This demonstrates the system's application in recognizing vehicles parked on the charging lane lines at a charging station and its effectiveness. Figure 6 The demonstration showcased the identification scenarios and effectiveness of illegally parked vehicles with blue license plates in charging station spaces. Analysis results demonstrate that this method can effectively and intelligently identify non-standard parking behaviors such as vehicles crossing the lines and illegally parking with blue license plates in charging spaces. Furthermore, it allows for the setting of early warning information based on the system's front-end configuration. Therefore, this system meets the requirements for automated identification and intelligent video inspection of non-standard parking behaviors in charging spaces and can be replicated and promoted in locations with charging spaces, such as shopping malls, office buildings, and residential communities.

[0056] In one embodiment, a method for recognizing parking regulations in charging spaces based on object detection and prior knowledge of coordinate geometry is provided, including: Obtain the camera configuration and perform timed frame extraction on the camera's video stream according to the inspection strategy to obtain the extracted frame images; Perform image correction and image enhancement processing on the extracted frames; The processed frame-segmented image is input into the target detection model to obtain target detection results that include vehicle targets and license plate targets. Read the polygon coordinate information and coordinate geometry determination rules of the corresponding parking space; Based on the target detection results and prior knowledge of coordinate geometry, the spatial relationship between the vehicle center point or license plate center point and the parking space area is determined in order to identify vehicle parking behavior that crosses the line and / or illegal parking behavior of blue-plate vehicles. Event logs are generated when irregular parking behavior is detected.

[0057] In this embodiment, the step of obtaining camera configuration and performing timed frame extraction on the camera's video stream according to the inspection strategy to obtain the extracted frame image includes: dividing the camera into a camera group for inspecting vehicles crossing the line and a camera group for inspecting illegal parking of blue-plate vehicles according to the camera's resolution, frame rate, installation angle and orientation, and configuring inspection cycles for each group.

[0058] In this embodiment, the image processing in the image correction and image enhancement processing of the frame-dropped image includes performing distortion correction, perspective correction, brightness enhancement, and contrast enhancement on the frame-dropped image.

[0059] In this embodiment, determining the spatial relationship between the vehicle center point or license plate center point and the parking space area includes: calculating the vehicle center point or license plate center point based on the target bounding box coordinates, and using the ray casting method to determine whether the center point is located inside the polygon of the parking space area.

[0060] In this embodiment, the method further includes: when the center point of the vehicle is located inside the parking space area, calculating the distance from the center point of the vehicle to each side of the parking space and selecting the minimum distance, comparing it with a preset distance threshold to determine whether the vehicle meets the line crossing condition.

[0061] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A charging parking space parking regulation recognition system based on target detection and coordinate geometry prior knowledge, characterized in that, The system includes: a video inspection strategy module, a target detection model module, and an irregular parking recognition module; The video inspection strategy module is used to manage the access of cameras corresponding to charging spaces, establish the association between cameras and inspection tasks, and perform timed frame extraction on the video stream according to a preset inspection cycle to obtain frame images. The target detection model module is used to perform target detection on the frame-sampling image to obtain target detection results including vehicle targets and license plate targets. The target detection results include target category labels and bounding box coordinates. The irregular parking identification module is used to: analyze the spatial relationship between the vehicle center point or license plate center point and the parking space area based on the target detection results and the coordinate geometry prior knowledge base, so as to identify the vehicle parking behavior of crossing the line and / or the illegal parking behavior of blue-plate vehicles, and generate an event record with event information. The coordinate geometry prior knowledge base is used to store the polygon coordinate information of the parking space area and the coordinate geometry rules used for judging irregular parking behavior.

2. The system according to claim 1, characterized in that, The video inspection strategy module includes a camera adaptation unit and a task configuration unit. The camera adaptation unit is used to obtain the resolution, frame rate, installation position and orientation of the camera and divide the camera into different inspection task groups according to preset rules. The task configuration unit is used to set the inspection time interval for different inspection task groups.

3. The system according to claim 1, characterized in that, The video inspection strategy module also includes an image processing unit, which performs distortion correction, perspective correction, brightness enhancement or contrast enhancement processing on the extracted frame images and generates image tags for identification and association.

4. The system according to claim 1, characterized in that, The target detection model module adopts a lightweight multi-scale target detection model and is configured to output the category labels and corresponding bounding box coordinates of vehicle targets, blue license plate targets, and green license plate targets.

5. The system according to claim 1, characterized in that, The non-standard parking recognition module is used to determine whether the center point of the vehicle or the center point of the license plate is located inside the polygonal area of ​​the parking space using the ray method, and calculates the minimum distance between the vehicle's center point and each side of the parking space when the center point of the vehicle is located inside the area to determine whether it is over the line.

6. A method for recognizing parking regulations in charging spaces based on target detection and coordinate geometry prior knowledge, characterized in that, include: Obtain the camera configuration and perform timed frame extraction on the camera's video stream according to the inspection strategy to obtain the extracted frame images; Perform image correction and image enhancement processing on the extracted frames; The processed frame-segmented image is input into the target detection model to obtain target detection results that include vehicle targets and license plate targets. Read the polygon coordinate information and coordinate geometry determination rules of the corresponding parking space; Based on the target detection results and prior knowledge of coordinate geometry, the spatial relationship between the vehicle center point or license plate center point and the parking space area is determined in order to identify vehicle parking behavior that crosses the line and / or illegal parking behavior of blue-plate vehicles. Event logs are generated when irregular parking behavior is detected.

7. The method according to claim 6, characterized in that, The process of obtaining camera configuration and performing timed frame extraction on the camera's video stream according to the inspection strategy to obtain frame-extracted images includes: dividing the cameras into camera groups for inspecting vehicles crossing the line and for inspecting illegally parked vehicles with blue license plates, based on the camera's resolution, frame rate, installation angle, and orientation, and configuring inspection cycles for each group.

8. The method according to claim 6, characterized in that, The image processing in the image correction and image enhancement process performed on the frame-stripped image includes distortion correction, perspective correction, brightness enhancement, and contrast enhancement.

9. The method according to claim 6, characterized in that, The determination of the spatial relationship between the vehicle center point or license plate center point and the parking space area includes: calculating the vehicle center point or license plate center point based on the target bounding box coordinates, and using the ray casting method to determine whether the center point is located inside the polygon of the parking space area.

10. The method according to claim 6, characterized in that, The method further includes: when the center point of the vehicle is located inside the parking space area, calculating the distance from the center point of the vehicle to each side of the parking space and selecting the minimum distance, comparing it with a preset distance threshold to determine whether the vehicle meets the line crossing condition.

Images