Parking lot monitoring method and device based on visual inspection and computer equipment

Abstract

The invention relates to a parking lot monitoring method and device based on visual inspection and computer equipment. The method comprises the following steps: acquiring a monitoring video corresponding to a parking lot according to a camera arranged in the parking lot; according to parking space image features, image pixel analysis is carried out on the monitoring video, the parking space occupation condition of each parking space in the parking lot is obtained, and feedback is carried out according to all parking space occupation conditions; and in combination with vehicle state characteristics and personnel activity characteristics, performing target detection analysis on the monitoring video to obtain vehicle safety conditions of each vehicle in the parking lot, and giving an alarm for abnormal vehicle safety conditions. By adopting the method, the comprehensive monitoring capability and the overall management capability of the running state of the parking lot can be realized.

Description

Visual detection-based parking lot monitoring methods, devices, and computer equipment Technical Field

[0001] This application relates to the field of parking lot monitoring technology, and in particular to a parking lot monitoring method, device and computer equipment based on visual detection. Background Technology

[0002] In the field of parking lot monitoring technology, traditional monitoring methods use cameras in conjunction with parking space sensors to detect parking space occupancy and assist in management. However, this type of method requires additional wiring during installation and maintenance, resulting in a complex structure, high construction costs, and significant maintenance burdens. Furthermore, sensors are susceptible to environmental factors such as temperature and humidity, leading to decreased detection accuracy and limiting the system's reliability and application scope. In addition, this method has relatively limited functionality and cannot meet the needs of modern parking lots for comprehensive management and intelligent identification. Summary of the Invention

[0003] Therefore, it is necessary to provide a parking lot monitoring method, device, computer equipment, and computer-readable storage medium based on visual detection to address the aforementioned technical problems.

[0004] Firstly, this application provides a parking lot monitoring method based on visual detection, comprising: acquiring monitoring video of the parking lot based on cameras deployed in the parking lot; performing image pixel analysis on the monitoring video based on parking space image features to obtain the parking space occupancy status of each parking space in the parking lot, and providing feedback on the occupancy status of all parking spaces; combining vehicle status features and personnel activity features to perform target detection analysis on the monitoring video to obtain the vehicle safety status of each vehicle in the parking lot, and issuing alarms for abnormal vehicle safety status.

[0005] Secondly, this application also provides a parking lot monitoring device based on visual detection, comprising: an acquisition module for acquiring monitoring video of the parking lot based on cameras deployed in the parking lot; an occupancy analysis module for performing image pixel analysis on the monitoring video based on parking space image features to obtain the parking space occupancy status of each parking space in the parking lot, and providing feedback on the occupancy status of all parking spaces; and a security analysis module for performing target detection analysis on the monitoring video by combining vehicle status features and personnel activity features to obtain the vehicle security status of each vehicle in the parking lot, and issuing alarms for abnormal vehicle security status.

[0006] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the above steps.

[0007] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the above steps.

[0008] The aforementioned visual detection-based parking lot monitoring method, device, computer equipment, and computer-readable storage medium, firstly, continuously acquire monitoring video from cameras deployed in the parking lot, forming a continuous, complete, and analyzable image sequence of the overall parking lot information. On one hand, by performing image pixel analysis based on parking space image features of the monitoring video, the occupancy status of each parking space can be accurately distinguished, thus providing a comprehensive display of the real-time usage of parking space resources. On the other hand, by performing target detection analysis combining vehicle status features and personnel activity features of the monitoring video, abnormal situations during vehicle parking can be identified in a timely manner, thereby forming effective monitoring of vehicle safety. Based on this, in the entire technical solution, through feature-level visual analysis of the monitoring video, parking lot management and vehicle safety monitoring form a unified and efficient visual recognition link, thereby achieving comprehensive monitoring and overall management capabilities for the parking lot's operational status. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 is a flowchart of a parking lot monitoring method based on visual detection in one embodiment; Figure 2 is a structural block diagram of a parking lot monitoring device based on visual detection in one embodiment. Detailed Implementation

[0011] 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.

[0012] In an exemplary embodiment, as shown in FIG1, a parking lot monitoring method based on visual detection is provided. This embodiment illustrates the method by applying it to a server. It is understood that the method can also be applied to a terminal, or to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps S101 to S103.

[0013] Step S101: Obtain the corresponding surveillance video of the parking lot based on the cameras installed in the parking lot.

[0014] For example, firstly, surveillance video covering the entire monitoring area is acquired using cameras pre-positioned in the parking lot; during the acquisition of surveillance video, the imaging parameters, monitoring angle, and frame rate range of the cameras need to be managed to ensure that the acquired surveillance video maintains stable image continuity and clarity, and avoids loss of image information due to changes in light, obstruction, or fluctuations in image quality.

[0015] Subsequently, the real-time video transmitted from the camera is fed into the corresponding processing terminal, transforming the surveillance video into an analyzable data stream. This stream includes the background of the parking space area, the outlines of vehicles, the dynamic changes of pedestrians as they pass by, and the image characteristics of various fixed structures in the environment. Next, this data stream is organized in a basic temporal sequence to ensure that each frame maintains the correct temporal relationship, thereby ensuring that subsequent analysis of dynamic changes is based on strict temporal continuity. Based on this, a preliminary image quality assessment is performed on the acquired surveillance video, appropriately processing or marking blurriness, jitter, or abnormal brightness in the image to provide a clean and analyzable image sequence for subsequent steps.

[0016] Through the above processing, structured surveillance video data is finally obtained. This data consists of image frames arranged in chronological order, which can accurately present the visual information of the parking lot at different points in time, providing reliable raw input for subsequent analysis of parking space occupancy and vehicle safety status.

[0017] Step S102: Based on the characteristics of the parking space image, perform image pixel analysis on the monitoring video to obtain the parking space occupancy status of each parking space in the parking lot, and provide feedback on the occupancy status of all parking spaces.

[0018] Among them, the parking space image feature represents the image attributes used to describe how the parking space is presented in the monitoring screen. It is used to distinguish whether the parking space is occupied by a vehicle or is vacant. For example, when the parking space is vacant, it is presented as a relatively complete ground texture, while when it is occupied by a vehicle, it is presented as a pixel structure change covered by the outline of the vehicle body.

[0019] For example, firstly, the image areas involving parking spaces in the surveillance video are located, so that the image areas corresponding to the parking space layout can be accurately divided, thereby forming image segments for each parking space. After the image area division is completed, pixel analysis of the image area content needs to be performed based on the characteristics of the parking space image. That is, by reading characteristics such as brightness changes, edge direction changes, and shape occupancy relationships, a distinguishable difference is formed between the pixel patterns belonging to vehicles and the pixel patterns in an empty state.

[0020] Next, these differences are mapped to the time dimension, and the pixel changes in consecutive frames are sorted out to obtain a state detection result that can represent the current state of each parking space. This state detection result can show whether each parking space is occupied or vacant.

[0021] Through the above process, a set of parking space occupancy information is finally obtained, which includes parking space number, parking time, and status detection results. This set can clearly show the actual usage of all parking spaces, so as to understand the distribution of parking space resources at different times.

[0022] Furthermore, in the parking space occupancy status set, vacant parking spaces will be clearly marked, thereby enabling the calculation of the number of currently available vacant parking spaces. Subsequently, the number of vacant parking spaces and their corresponding parking space numbers will be fed back to the preset parking management system in real time, for example, by sending them to the parking management system via a network communication protocol (such as TCP / IP protocol). This allows the management personnel to guide and dispatch parking spaces based on their distribution, while also allowing drivers to understand the parking space occupancy status before entering the parking lot, thereby improving overall parking efficiency.

[0023] Step S103: Combine vehicle status characteristics and personnel activity characteristics to perform target detection analysis on the monitoring video, obtain the vehicle safety status of each vehicle in the parking lot, and issue alarms for abnormal vehicle safety status.

[0024] Among them, vehicle status features represent visual attributes used to describe the appearance and position changes of a vehicle in the monitoring screen. They are used to determine whether a vehicle is stationary, moving, close to the parking space boundary, or in an abnormal state. For example, when the vehicle position remains unchanged in continuous images, it presents a stable outline, while when the vehicle moves, it presents an outline shift or shape change.

[0025] Among them, personnel activity features are used to describe the visual attributes of personnel's actions and trajectories in the monitoring screen. They are used to identify the approach, stay or abnormal behavior between personnel and vehicles or parking spaces. For example, when a person gradually approaches a vehicle from a distance, it is presented as a shortening spatial distance and a continuous displacement trajectory frame by frame.

[0026] For example, on the one hand, changes in the appearance of vehicles in the image are identified based on vehicle status characteristics, including changes in vehicle orientation, body outline, and headlight illumination, thereby enabling the identification of the vehicle's state at different points in time from continuous images. On the other hand, the dynamic trajectories of people in the image are identified based on human activity characteristics. By reading the relative distance between people and vehicles, their approach direction, and their stopping position, different human behaviors can be visually distinguished.

[0027] Next, the vehicle status characteristics are correlated with the personnel activity characteristics to understand the relationship between personnel behavior and vehicle status, such as approaching, stopping, illegal touching, or other abnormal behaviors. During this process, the dynamic changes of each frame in the surveillance video need to be sequentially analyzed to form a continuous chain of visual evidence for any abnormal behavior, thereby improving the reliability of the judgment.

[0028] Finally, the results of the above correlation analysis are compiled and unified to obtain a set of vehicle safety statuses that include the normal or abnormal states of each vehicle. When an abnormal state is identified, the corresponding parking space, vehicle identification, abnormal time, and abnormal type will be clearly marked to trigger the alarm process, that is, to feed back the relevant information to the preset parking management system, for example, by sending a warning SMS to the manager or vehicle owner through an SMS interface (such as using the API of a third-party SMS service provider), so that the manager or vehicle owner can be informed of the vehicle status in a timely manner and take appropriate measures.

[0029] Optionally, any of the above data processing procedures can be deployed on a server system equipped with multi-core CPUs, large-capacity memory, and high-performance GPUs, enabling various image processing and detection algorithms to maintain stable and efficient operation under high concurrency and high computational load conditions. Based on this, by executing parking space recognition, target detection, and anomaly analysis processes in a high-performance hardware environment, processing latency can be significantly reduced and overall response speed improved. Simultaneously, to ensure the accuracy of the system in actual use, the relevant algorithms need to be periodically optimized based on operational feedback. For example, in parking space recognition, the pixel change threshold may be recalibrated, or in anti-theft warning analysis, the detection time threshold may be adjusted, allowing the algorithms to better reflect changes in real-world scenarios. Furthermore, to ensure the continuous and reliable operation of the system, performance monitoring of server load, memory usage, and processing queues is necessary to maintain server stability when processing large-scale video streams, thereby supporting long-term and continuous monitoring services.

[0030] In the aforementioned visual detection-based parking lot monitoring method, in step S101, monitoring videos are continuously acquired by cameras deployed in the parking lot, forming a continuous, complete, and analyzable image sequence of the overall parking lot image information. In step S102, image pixel analysis based on parking space image features is performed on the monitoring videos, enabling accurate differentiation of the occupancy status of each parking space, thus providing a comprehensive display of the real-time usage of parking space resources. In step S103, target detection analysis combining vehicle status features and personnel activity features is performed on the monitoring videos, enabling timely identification of abnormal situations during vehicle parking, thereby forming effective monitoring of vehicle safety status. Based on this, in the entire technical solution, feature-level visual analysis of monitoring videos enables parking lot space management and vehicle safety monitoring to form a unified and efficient visual recognition link, thereby achieving comprehensive monitoring and overall management capabilities for the parking lot's operational status.

[0031] In an exemplary embodiment, the monitoring video corresponding to the parking lot is obtained based on the cameras deployed in the parking lot, including steps S201 to S202.

[0032] Step S201: Cameras are pre-positioned in the entrance / exit area, passage area, and parking space area of ​​the parking lot. The imaging parameters of the corresponding cameras are determined according to the scene characteristics of each area to obtain the area monitoring video of the parking lot in each area.

[0033] The entrance and exit area represents the functional space in the parking lot used for vehicles or people to enter and exit; the passage area represents the path space inside the parking lot used for vehicles to drive or people to pass; and the parking space area represents the fixed space in the parking lot used for vehicles to park.

[0034] Among them, the scene characteristics of the area represent the typical visual attributes of each area in terms of structural layout and screen content. For example, the entrance and exit area presents obvious entrance and exit signs, the parking space area presents regularly arranged parking lines, and the passage area presents a long extension.

[0035] For example, firstly, to ensure sufficient scene coverage and imaging consistency in the monitoring images of different areas of the parking lot during subsequent processing, cameras need to be pre-positioned in the entrance and exit areas, passage areas, and parking space areas. During this process, based on the functional differences of each area in actual operation, the spatial structure, lighting distribution, and flow direction of people and vehicles in each area are observed, so that the installation position of each camera can obtain images representing the typical state of that area.

[0036] Subsequently, the imaging parameters of the cameras are set according to the scene characteristics of these areas, ensuring that their focal length, imaging angle, image coverage, and brightness response are consistent with the monitoring needs of the corresponding areas, thereby generating clear and recognizable image content during the acquisition process. Next, with multiple cameras operating simultaneously, the entrance / exit area is used to display images of vehicles entering and exiting, the passageway area is used to display images of vehicles moving and people moving, and the parking space area is used to display images of parking space occupancy, enabling stable generation of area monitoring videos for each area. Furthermore, to ensure the temporal continuity of the image content, preliminary temporal organization is required for the image sequences captured by the cameras, ensuring that the images from different cameras at the same time maintain a consistent time signature.

[0037] Through the above process, we finally obtain regional surveillance videos with clear regional attributes, which are collected by multiple cameras. These regional surveillance videos can clearly present the visual changes of each area in the parking lot spatial structure, providing accurate and sectionally callable basic data for subsequent image pixel analysis and target detection analysis.

[0038] Step S202: Classify and store the regional surveillance videos of each area according to the multi-dimensional acquisition attributes to obtain surveillance videos represented by the video index structure, which can be used to call the matching regional surveillance videos before image pixel analysis and target detection analysis.

[0039] Among them, multidimensional acquisition attributes represent the combined characteristics of surveillance video in terms of regional attributes, time attributes, and image change attributes, which are used to classify and manage video content from different sources.

[0040] For example, firstly, to enable subsequent analysis processes to retrieve the corresponding regional surveillance videos according to different processing needs, the regional surveillance videos acquired in the preceding steps need to be classified and stored according to multi-dimensional acquisition attributes. During this process, based on the regional attributes, time attributes, and image structure attributes possessed by the regional surveillance videos during acquisition, they are tagged so that each video segment can be clearly identified in the storage system with its acquisition attributes.

[0041] Subsequently, these videos are categorized by tags, and area surveillance videos under the same attribute category are organized into continuously searchable data sets, thus forming a categorized storage structure that reflects the spatial structure and temporal changes of the parking lot. Further, after classification, an indexable management structure, i.e., a video index structure, needs to be generated for different categories of videos, enabling area surveillance videos to be retrieved through a fixed indexing method. This video index structure must be able to distinguish between area types, time segments, and image change characteristics, allowing for rapid location of the corresponding video set according to specific needs during image pixel analysis or target detection analysis.

[0042] Through the above process, the surveillance video represented by the video index structure is finally obtained. This surveillance video not only retains the original image content of the regional surveillance video, but also has the retrieval capability that can be called on demand, providing an accurate video source for image pixel analysis and target detection analysis, so that the subsequent analysis process can be carried out within the correct area.

[0043] In this embodiment, in step S201, cameras are deployed in the entrance / exit area, passage area, and parking area respectively, and imaging parameters are set according to the scene characteristics of each area to obtain regional monitoring videos. This ensures that the image content of each area has clear imaging performance that meets the monitoring requirements, thus providing a stable and distinguishable visual foundation for subsequent video retrieval and analysis of different areas. In step S202, the regional monitoring videos are classified and stored according to multi-dimensional acquisition attributes and organized with a video index structure, so that different categories of regional monitoring videos can be clearly identified and quickly retrieved, thereby ensuring that regional monitoring videos matching the processing target can be retrieved before analysis. Based on this, in the entire technical solution, through regional acquisition and structured management, the monitoring video forms a clear transmission link from acquisition to storage to retrieval, thereby providing accurate, controllable, and efficient video data support for subsequent image pixel analysis and target detection analysis.

[0044] In an exemplary embodiment, image pixel analysis is performed on the surveillance video based on the parking space image features to obtain the parking space occupancy status of each parking space in the parking lot, including steps S301 to S302.

[0045] Step S301: Based on the boundary features in the parking space image features, perform edge extraction and edge enhancement on the monitoring video based on the parking space boundary to obtain the target image area of ​​each parking space in the parking lot in the monitoring video.

[0046] Among them, the boundary features in the parking space image feature represent the outline information of the parking space formed by fixed lines or structures in the image, so as to determine the actual range of the parking space in the image. For example, the parking space ground markings appear as bright or continuous edge lines in the image.

[0047] For example, firstly, in order to make the geometric outline of the parking space in the image highly recognizable, it is necessary to read out the area with obvious structural lines in consecutive image frames based on the fixed structural characteristics of the parking space ground markings, and to maintain temporal consistency with the original image during the extraction process.

[0048] Subsequently, by enhancing the edges of these structural lines, the outer contours of the parking spaces can form clearer boundaries in the image, thereby reducing the interference of the complex background of the parking lot on the delineation of parking space areas. In the image after edge enhancement, the area of ​​the parking space can be gradually converged into an image region with identifiable boundaries.

[0049] Next, the monitoring screen is divided into regions based on the image regions after boundary extraction, so that each parking space can be mapped to an independent target image region. During the division process, the spatial position of the parking space boundary appearing in each frame is fixed, so that the difference between the parking space ground markings and the surrounding background is stably mapped to the corresponding region. Thus, the final target image regions have stable boundaries, fixed positions, and independent image ranges.

[0050] Step S302: Based on the pixel change features in the parking space image features, perform image pixel analysis on each target image area in the monitoring video to obtain the parking space occupancy status of each parking space in the parking lot.

[0051] Among them, the pixel change feature in the parking space image features represents the image attribute used to describe the change of pixel value over time and the change of pixel value with position within the target image area, in order to determine whether the parking space is vacant or occupied by a vehicle. For example, after a vehicle parks in the parking space, the originally uniform ground texture pixels are covered by the vehicle surface pixels, forming obvious brightness or texture changes.

[0052] For example, firstly, in order for pixel changes within the target image area to reflect whether a vehicle occupies a parking space, it is necessary to read the brightness changes, color coverage relationships, and structural texture changes in consecutive frames, so that the pixel differences presented by the original ground texture after the vehicle parks can be clearly captured. Subsequently, these changes are processed in both temporal and spatial dimensions, so that the spatial pixel distribution changes within each frame and the pixel change trends between adjacent frames can be presented simultaneously, thereby determining whether the target image area has undergone a state transition from vacant to occupied or from occupied to vacant in consecutive images.

[0053] Next, the pixel changes within the target image area are combined with the fixed boundaries of the parking space to provide a stable spatial reference for identifying occupancy status, avoiding misjudgments caused by external lighting or image noise. Furthermore, based on the processed pixel changes, a corresponding analysis result is generated for each target image area, allowing the analysis result to be identified as occupied or vacant, and establishing a one-to-one correspondence with the parking space number or location.

[0054] Throughout the analysis, continuous reading of the region's pixel structure allows for state comparison and determination of each parking space within a unified time and spatial scale. The final result is the parking space occupancy status of each parking space within the parking lot, clearly reflecting the actual usage of each parking space in the surveillance video.

[0055] Optionally, each frame of the image is first preprocessed, such as grayscale conversion and noise reduction, to improve image quality. Then, an edge detection algorithm is used to extract the parking space boundary to determine the parking space area. The pixel changes in the parking space area are compared in multiple consecutive frames to reveal the texture differences between vacant and occupied parking spaces. Based on the magnitude and pattern of pixel changes, it can be determined whether the parking space is covered by a vehicle. Further, the above process can be implemented using the OpenCV library in Python. For example, grayscale conversion and Gaussian filtering can be used for preprocessing. Then, Canny edge detection and morphological operations can be used to determine the parking space area. A custom pixel change analysis function can be used to compare the pixel differences in consecutive frames. For example, the average difference of pixel values ​​within the parking space area over multiple frames can be calculated. When the difference is consistently below a threshold, the space is considered vacant; when it is consistently above the threshold and exhibits vehicle shape characteristics, it is considered occupied.

[0056] In this embodiment, in step S301, edge extraction and edge enhancement are performed on the monitoring video based on the boundary features in the parking space image features to obtain the target image area corresponding to each parking space, so that the actual range of the parking space in the monitoring screen is clearly defined. In step S302, image pixel analysis is performed on the target image area based on the pixel change features in the parking space image features, so that the image changes in the target image area can be accurately identified, thereby making the parking space occupancy status clearly distinguishable. Based on this, in the entire technical solution, through the area positioning based on the parking space boundary and the two-dimensional analysis based on pixel change, the parking space occupancy status can be reliably judged within a precise area, thereby achieving the ability to accurately identify the parking space occupancy status.

[0057] In an exemplary embodiment, based on the pixel change features in the parking space image features, image pixel analysis is performed on each target image region in the monitoring video to obtain the parking space occupancy status of each parking space in the parking lot, including step S401.

[0058] Step S401: Based on the pixel change features in the parking space image features, perform parking scene correlation analysis on the multi-dimensional pixel change patterns of each target image region in the monitoring video to obtain the comprehensive change information corresponding to each target image region, which can be used as the parking space occupancy status of each parking space in the parking lot.

[0059] Among them, the multi-dimensional pixel change mode represents the pixel change pattern in the target image area composed of both time and space dimensions. It is used to reflect the comprehensive changes in brightness, texture and coverage structure of parking spaces in continuous monitoring images. For example, in the time dimension, it presents the change of pixel values ​​being covered by vehicles frame by frame, and in the space dimension, it presents the change of the transition from ground texture to vehicle texture within the area.

[0060] For example, firstly, in order to obtain effective pixel change information that can reflect vehicle parking behavior, it is necessary to read the brightness differences, ground texture changes, and coverage relationship changes in consecutive frames, so that the pixel change trend of each target image area in the time and space dimensions can be fully expressed.

[0061] Subsequently, these pixel change trends are organized, mapping changes in the time series, local changes within the spatial region, and changes near the boundaries into a unified multi-dimensional pixel change pattern. This allows different types of image changes to be mapped to a unified analytical framework. Next, to establish a correspondence between the multi-dimensional pixel change pattern and actual parking scenarios, the regional change information needs to be correlated with factors such as the background structure of the parking space, changes in coverage position after a vehicle parks, and changes in ambient lighting. This ensures that the interpretation of pixel changes can be reasonably explained within the overall parking lot scene.

[0062] Through the aforementioned parking scene correlation analysis, core change features reflecting changes in parking space status can be extracted from the original pixel differences, enabling the pixel behavior of each target image region at different time points to form describable structured image change information. Furthermore, by integrating this structured image change information, comprehensive change information is generated for each target image region. This comprehensive change information can directly express the actual state of the target image region in the surveillance video, accurately reflecting the change in parking space status from vacant to occupied or from occupied to vacant.

[0063] For example, in the target image area corresponding to a parking space, the brightness of the ground texture gradually decreases in consecutive frames, while the dark pixels formed by the vehicle's appearance expand towards the center of the area frame by frame. This change manifests as deepening coverage in the temporal dimension and pixel replacement advancing from the boundary in the spatial dimension. Organizing this pixel change trend forms a multi-dimensional pixel change pattern that includes both temporal and spatial relationships. This multi-dimensional pixel change pattern is then correlated with the actual parking lot scene. For instance, the area originally had clear parking space markings, but these markings are obscured after a vehicle is present; or the lighting remains stable over the same period, allowing the pixel changes to be attributed to vehicle entry rather than light fluctuations. Therefore, through this parking scene correlation analysis method, the source of pixel changes can be reasonably explained, thus determining whether the parking space has changed from vacant to occupied.

[0064] In this embodiment, in step S401, the pixel change features in the parking space image features of the monitoring video are organized so that the image information of the target image area can be uniformly expressed as a multi-dimensional pixel change pattern. Furthermore, the multi-dimensional pixel change pattern is subjected to parking scene correlation analysis so that the source of pixel change can be reasonably explained in the actual parking scene, thereby providing a clear scene basis for judging the parking space occupancy status. Based on this, in the entire technical solution, by abstracting the original pixel changes into comprehensive change information with scene correlation, the process of identifying parking space occupancy status is improved from local pixel judgment to multi-dimensional correlation judgment, thereby achieving a more accurate and robust parking space occupancy detection capability.

[0065] In an exemplary embodiment, the surveillance video is analyzed by combining vehicle status characteristics and personnel activity characteristics to obtain the vehicle safety status of each vehicle in the parking lot, including steps S501 to S503.

[0066] Step S501: Based on the vehicle status characteristics, perform vehicle target detection analysis on the monitoring video to obtain the vehicle detection results of each vehicle during the parking process.

[0067] For example, firstly, to ensure that vehicles can be identified under different lighting conditions and angles, it is necessary to read the stable structural features of the vehicle's appearance from consecutive frames of the surveillance video, such as the body outline, roof shape, headlight structure, and overall volume, so that the vehicle can form a resolvable difference from the background environment in the image. Subsequently, by identifying the edge shape, structural distribution, and coverage area of ​​the vehicles in the image, the position and appearance of each vehicle in the image are accurately mapped to the visual area corresponding to the vehicle target, so that the spatial position and shape features of the vehicle in the surveillance image are clearly presented. Here, the visual area represents the recognizable spatial range composed of image content in the surveillance image, which is used to carry the position and appearance information of the vehicle target in the image. For example, the rectangular image area occupied by the vehicle in the surveillance video can be used to describe the specific position of a particular vehicle in the image coordinate system.

[0068] Next, to ensure accurate tracking of vehicle state changes across different frames, it is necessary to organize the vehicle's positional and component state changes over time. This includes recording situations such as the vehicle remaining stationary, moving at low speed, or undergoing posture changes, as well as whether headlights are on / off or windows are open / closed, in the vehicle detection results. This ensures that the vehicle's state at different points in time is clearly represented. Furthermore, the spatial relationship between the vehicle and its surrounding environment must be read during the recognition process. This allows information such as whether the vehicle is within a normal parking range, whether it has deviated from parking lines, or whether it exhibits abnormal movement trends in different areas to be incorporated into the vehicle detection results.

[0069] For example, in a surveillance video of a parking space, consecutive frames show a white vehicle gradually entering the frame, its outline maintaining a stable edge structure under different lighting conditions. Based on these edge shapes and structural distributions, the rectangular area occupied by the vehicle in the image is defined as the corresponding visual region, accurately representing the vehicle's position and appearance in the image coordinates. Subsequently, by reading the changes in the vehicle's position within this visual region in consecutive frames, it can be observed that the vehicle gradually changes from slowly moving forward to coming to a standstill. Combined with the parking space boundaries, it is possible to further determine whether the vehicle is accurately parked in the space; for example, if the vehicle deviates from the lane markings, this deviation will be recorded in the vehicle detection results.

[0070] Step S502: Based on the characteristics of personnel activity in each vehicle within its corresponding physical range, perform personnel target detection analysis on the monitoring video to obtain the personnel detection results for each vehicle within its corresponding physical range.

[0071] The physical range corresponding to the vehicle refers to the actual activity monitoring area defined by the vehicle's position in the monitoring screen. It is used to limit the spatial boundary of personnel detection and analysis. For example, the area around the vehicle extends outward from the vehicle's outline and can be used to identify personnel behaviors related to the vehicle, such as approaching, stopping, or detouring.

[0072] For example, firstly, in consecutive frames, the physical structure of a person is extracted, such as body contours, movement trajectories, and posture changes, so that the person's activities around the vehicle can form a relatively independent visual representation from the background environment. Then, to ensure that the judgment of vehicle safety status has a clear physical scope, the analysis area of ​​personnel activity needs to be defined according to the surrounding area of ​​each vehicle. The position and physical structure of the person entering this physical scope are then mapped to the visual area corresponding to the person target, so that the position and physical structure of the person in the monitoring screen are clearly presented. Here, the visual area represents the identifiable spatial range composed of image content in the monitoring screen, used to carry the position and appearance information of the person target in the image. For example, the rectangular screen area occupied by a person in the monitoring video can be used to describe the specific position of a particular person in the image coordinate system.

[0073] Next, the behavioral and trajectory information of the person within the corresponding physical range is read, such as behavioral information like opening a door handle or observing a car window, and trajectory information such as approaching a vehicle, moving around a vehicle, or stopping briefly or continuously, so that the changes in the person's actions within the physical range can form a complete time series record.

[0074] For example, in a surveillance video, a person enters the physical area corresponding to a vehicle from the direction of the passage. Their body outline is clearly extracted in consecutive frames, and their movement trajectory shows a path of gradually approaching the vehicle. Their position and appearance in the image are projected onto the visual area corresponding to the person, so that the person's position and appearance in the image coordinates are accurately expressed. Then, the trajectory changes and behavioral changes of the person in this visual area are read in consecutive frames. When the person approaches the vehicle, they first pause briefly near the door, then reach out to touch the door handle and observe the window. Their actions form a complete time series record in consecutive frames.

[0075] Furthermore, to ensure the interpretability of behavioral changes, the analysis should incorporate the duration of a person's stay at different locations, their direction of movement, and their proximity. This allows the activity of people in the footage to correspond to the location of vehicles, enabling a correct distinction between approaching behavior and simply passing by, thus ensuring that the personnel detection results have a clear behavioral nature. For example, in surveillance footage, if a person gradually approaches a parked vehicle from the direction of the passageway, lingers around the vehicle for a relatively long time, consistently moves towards the vehicle, and then slowly moves along one side of the vehicle after approaching, this behavior can be determined as an approach directly related to the vehicle, rather than simply passing by. Conversely, if another person merely walks along the passageway in a straight line without stopping near the vehicle, this is identified as simply passing by.

[0076] Through the above processing, the final result is the personnel detection results of each vehicle within its corresponding physical range. These personnel detection results can record the behavioral and trajectory information of personnel around the vehicle, providing a reliable data basis for subsequent combined analysis with vehicle detection results.

[0077] Step S503: Combine the vehicle detection results and personnel detection results corresponding to the same vehicle to obtain the vehicle safety status of the corresponding vehicle, so as to integrate the vehicle safety status of each vehicle in the parking lot.

[0078] For example, firstly, the vehicle detection results include not only changes in the vehicle's position and status within the monitoring screen, but also changes in the status of the vehicle's external components within consecutive images, such as the on / off state of headlights, the open / closed state of windows, or changes in the position of doors. This ensures that the vehicle's stationary, moving, or component changes at different points in time can be completely recorded. Furthermore, to provide a clear analytical reference for these vehicle status changes, the corresponding image range mapped to the vehicle's physical area in the image needs to be used as a correlation benchmark, allowing personnel detection results to be matched according to this image range.

[0079] Subsequently, the trajectory and behavioral changes of personnel included in the personnel detection results need to be correlated with the image range corresponding to the vehicle, so that personnel activities such as approaching, stopping, going around, and touching the vehicle can establish a temporal and spatial correspondence with changes in the vehicle's state. Then, by comparing the correlation between the two, such as whether personnel approached before or after abnormal changes in vehicle parts, or whether there were any touching actions on the windows or doors when the vehicle was stationary, the combination relationship between the vehicle detection results and the personnel detection results can form a continuous event chain.

[0080] Finally, by integrating the vehicle inspection results with the personnel inspection results, the vehicle safety status is obtained, which can accurately reflect the comprehensive situation of the vehicle in terms of its location, component status, and the activities of people in the surrounding area.

[0081] Optionally, a YOLO detection model for vehicles and people can be built using the deep learning framework TensorFlow. A large number of images containing vehicles and people can be used as training data. By labeling the location ranges of the images, the model can learn the appearance features of vehicles and people and their typical shapes in the image. Subsequently, the labeled data is input into the YOLO detection model for training. By setting parameters such as the learning rate, number of training epochs, and batch size, the model continuously improves its target recognition ability through multiple iterations until it can stably output the location results of vehicles and people. After training, monitoring video frames can be read in real time, and each frame can be input into the trained YOLO detection model. The model can then quickly return the detection boxes and their confidence scores for vehicles and people in the image, thereby achieving real-time recognition of dynamic targets in the monitoring footage.

[0082] In this embodiment, in step S501, the monitoring video is analyzed based on vehicle target detection according to vehicle state characteristics to obtain vehicle detection results for each vehicle during parking, thus providing a structured visual basis for judging the vehicle's own state; in step S502, the monitoring video is analyzed based on personnel target detection according to the personnel activity characteristics of each vehicle within its corresponding physical range to obtain personnel detection results for each vehicle within its corresponding physical range, thus providing a stable behavioral basis for judging the impact of personnel activities on vehicles; in step S503, the association processing of the combination relationship between vehicle detection results and personnel detection results forms a traceable event chain between changes in vehicle state and personnel activities, thereby enabling a clear judgment logic for vehicle safety status; based on this, in the entire technical solution, through bidirectional association analysis of vehicle state and personnel activities, vehicle safety monitoring is upgraded from single identification to multi-factor comprehensive judgment, thereby achieving a more reliable vehicle safety status identification capability.

[0083] In an exemplary embodiment, the vehicle safety status of the corresponding vehicle is obtained by combining the combined relationship between the vehicle detection results and the personnel detection results corresponding to the same vehicle, including steps S601 to S603.

[0084] Step S601: From the vehicle detection results and personnel detection results corresponding to the same vehicle, obtain the vehicle feature elements of the vehicle detection results and the personnel feature elements of the personnel detection results.

[0085] Among them, vehicle feature elements represent the structured information used to describe the vehicle status in the vehicle detection results, that is, reflecting the changes in the vehicle's position and component status in the monitoring screen, such as whether the vehicle is stationary or moving in the screen, the on / off status of the headlights, or the open / closed status of the windows.

[0086] Among them, personnel characteristic elements represent the structured information used to describe personnel activities in the personnel detection results, that is, reflecting the trajectory changes and behavioral changes of personnel within the physical range corresponding to the vehicle, such as approaching the vehicle, staying near the door, observing the window, or touching the door handle.

[0087] Step S602: Combine each vehicle feature element with each person feature element according to the relationship between the feature elements to obtain multiple feature element combinations.

[0088] Among them, the feature element combination represents a group of information formed by vehicle feature elements and personnel feature elements according to their relationship. It is used to express the correspondence between vehicle status changes and personnel activities within a certain time period, such as the feature combination formed when the vehicle remains stationary and the personnel stay in the door area.

[0089] For example, firstly, the changes in vehicle position and component status included in the vehicle feature elements need to correspond to the changes in trajectory and behavior included in the personnel feature elements, so that the logical relationship between different types of features can be reflected in the combined structure. In this process, in order to make these logical relationships have a clear structure, it is necessary to match the changes in vehicle status within the corresponding time period according to the location and time sequence of personnel activities, so that the changes in vehicle position and component status are naturally associated with personnel activities.

[0090] Next, multiple logically related feature elements are constructed into a feature element combination, such that each feature element combination contains at least one vehicle feature element and one person feature element. For example, when a vehicle is stationary and a person remains continuously around the vehicle, or when a vehicle part changes and a person makes contact nearby, these feature element combinations can clearly express the correlation of events.

[0091] Subsequently, in order to ensure the integrity of the combined structure, it is necessary to include the feature elements under different time periods and different positional relationships in the combination process, so that all feature elements that may constitute the meaning of the event are integrated into different feature element combinations.

[0092] Step S603: Based on the risk attributes reflected by each combination of feature elements, perform joint analysis on multiple combinations of feature elements to obtain the vehicle safety status of the corresponding vehicle.

[0093] Among them, the risk attribute represents the risk level of each combination of feature elements in terms of the meaning of the event, and is used to reflect the nature of the behavior indicated by the combination of feature elements, such as low-risk viewing behavior, medium-risk abnormal lingering behavior, or high-risk contact with the car door behavior.

[0094] For example, firstly, the correlation between vehicle and personnel features within each feature combination can reflect different event meanings, such as prolonged proximity to a stationary vehicle, personnel contact before and after changes in vehicle component states, and the correspondence between changes in vehicle position and absence of personnel behavior. To ensure the accurate understanding of the event meanings expressed by these feature combinations, their risk attributes need to be determined, allowing each feature combination to be logically categorized as normal behavior, suggestive behavior, or abnormal behavior.

[0095] Subsequently, joint analysis is performed among multiple combinations of feature elements, integrating feature element combinations from different time periods and location relationships. This allows the vehicle's status and the activities of surrounding personnel throughout the monitoring process to form an interpretable event chain. Through this event chain, the interaction between the vehicle and personnel can be extended from a single event to a continuous event, thus providing a complete temporal logic and spatial reference for judging the vehicle's safety status. For example, in the monitoring footage, one set of feature element combinations shows that the vehicle remains stationary and the headlights do not change; another set of feature element combinations shows that a person lingers in the door area multiple times with touching actions; and a subsequent set of feature element combinations shows that the window opens and closes within a short period of time. Thus, by jointly analyzing these feature element combinations from different time periods, an event chain containing "continuous approach - touching the door - window change" can be formed.

[0096] Next, it is necessary to process the conflicting, complementary, or continuous relationships between combinations of feature elements so that the final vehicle safety status can reflect both the meaning of a single event and the combined impact of multiple events. For example, in a conflicting relationship, one combination of feature elements shows that the vehicle has slightly shifted its position in a short period of time, while another combination shows that no one is approaching. The two points are inconsistent, and further comparison and verification are needed to determine whether there is external interference. In a complementary relationship, one combination of feature elements shows that people are staying in the door area, while another combination shows that the window is subsequently opened. These two sets of information complement each other and can both point to the vehicle being affected by external actions. In a continuous relationship, one combination of feature elements shows that people are gradually approaching the vehicle, another combination shows that people are staying by the side of the vehicle, and then the next combination shows that the vehicle parts change accordingly. Multiple sets of information form a coherent chain in time sequence, which can fully represent the continuous interaction between the vehicle and the people.

[0097] Based on this, the vehicle safety status of the corresponding vehicle is finally obtained. This vehicle safety status is formed by the joint analysis of the combination of feature elements, which can reflect the safety status of the vehicle in the current monitoring process and provide a directly usable structured judgment result for parking lot safety management.

[0098] In this embodiment, in step S601, various feature elements extracted from vehicle detection results and personnel detection results are used to form well-structured basic data of vehicle status changes and personnel activity information, thereby establishing a clear source of features for subsequent correlation judgment. In step S602, the correlation between different feature elements is combined to explicitly express the correspondence between vehicle status and personnel activity, thereby forming multiple feature element combinations that can be used for risk judgment. In step S603, the risk attributes presented by multiple feature element combinations are jointly analyzed to integrate vehicle status and personnel activity in the entire monitoring process into a causally related event chain, thereby generating a vehicle safety status with complete logical basis. Based on this, in the entire technical solution, through the layer-by-layer analysis of feature extraction, feature combination, and event correlation, vehicle safety judgment is improved from single phenomenon identification to multi-factor cross-verification, thereby achieving a more accurate vehicle safety status identification capability.

[0099] In an exemplary embodiment, based on the risk attributes reflected by each combination of feature elements, a joint analysis of multiple feature element combinations is performed to obtain the vehicle safety status of the corresponding vehicle, including steps S701 to S702.

[0100] Step S701: According to the risk assessment rules preset based on the parking scenario, perform risk attribute analysis on each combination of feature elements to obtain the risk attribute data corresponding to each combination of feature elements.

[0101] Among them, the risk assessment rules based on the parking scenario are the judgment criteria established according to the actual operating environment of the parking lot to judge the risk level of personnel activities and vehicle status, so as to classify risks based on different combinations of characteristic elements.

[0102] Among them, risk attribute data represents structured risk information obtained after analyzing the combination of characteristic elements according to risk assessment rules, and is used to characterize the specific attributes of the combination of characteristic elements in terms of risk level, behavior nature or event category.

[0103] For example, firstly, in order to ensure that the risk analysis process has a unified judgment basis, it is necessary to summarize the event meanings of different types of characteristic elements in the activity scenarios around the vehicle in advance, and form risk assessment rules for judging risks. That is, risk assessment rules are usually set based on changes in vehicle status, changes in personnel activities and the correspondence between the two. For example, continuous approach, contact with vehicle parts or sudden changes in vehicle parts can all be given clear risk interpretations in the rule system.

[0104] Subsequently, each combination of feature elements is input into the risk assessment rules. This allows for the comparison of vehicle status changes, personnel activity changes, and the corresponding relationships within each feature element combination according to the risk assessment rules, ensuring that each feature element combination corresponds to a specific risk attribute. Next, to ensure the risk attributes are recordable, the analyzed risk attributes need to be transformed into risk attribute data, enabling structured descriptions of different feature element combinations in terms of risk level, behavioral nature, or event category.

[0105] For example, in surveillance footage, a combination of features shows a vehicle remaining stationary while a person repeatedly lingers near the door and touches it. After comparison, this is classified as high-risk, with the behavior type being "abnormal contact" and the event category being "potential damage." As another example, another combination of features shows a person only briefly stopping beside the vehicle without touching it. After comparison, this is classified as low-risk, with the behavior type being "brief observation" and the event category being "normal passage."

[0106] Step S702: Associate the risk attribute data corresponding to the same combination of feature elements, and aggregate the risk attribute data corresponding to different combinations of feature elements to obtain the vehicle safety status of the corresponding vehicle.

[0107] For example, firstly, for risk attribute data generated by the same combination of feature elements, attribute association is performed based on their consistency in risk level, behavior nature, or event category, so that the internal logic of the risk expression of the feature element combination can maintain integrity in the analysis process; thus, through this attribute association method, the risk attribute data within the feature element combination can form a unified risk meaning, making the risk expression of the vehicle in a certain continuous behavior more concentrated.

[0108] For example, during the monitoring of a vehicle, a combination of feature elements shows that a person repeatedly lingers near the vehicle door while the vehicle remains stationary and the window position does not change. According to the risk assessment rules, this combination of feature elements is labeled as risk attribute data of "medium risk level - abnormal lingering - suspicious approach". Since these risk attribute data all point to the same type of risk meaning, they need to be grouped into the same risk expression through attribute association to ensure consistency during the analysis process. Thus, "medium risk level - abnormal lingering - suspicious approach" is associated with the unified risk interpretation of "abnormal approach to the vehicle". This allows the risk attributes presented by this combination of feature elements in different dimensions to be merged into a centralized expression, giving the risk status of the vehicle at this stage of behavior a clear, centralized, and traceable meaning.

[0109] Subsequently, the risk attribute data generated by different combinations of feature elements are aggregated. Based on the continuity between feature element combinations in terms of risk level, behavioral nature, or event category, they are integrated to form an event structure with temporal logic and spatial correlation. For example, if several combinations of feature elements show progressive changes in risk attributes, they need to be aggregated into a continuous event to avoid risk fragmentation caused by single behavioral judgments.

[0110] For example, during a monitoring period, different combinations of feature elements generated multiple risk attribute data: the first group showed a person approaching the vehicle's side door and lingering briefly, labeled "low-risk level - brief observation - normal approach"; the second group showed a person lingering in the same location again for an extended period, labeled "medium-risk level - abnormal lingering - suspicious approach"; and the third group showed a person touching the door handle, labeled "high-risk level - contact with the door - potential damage". These risk attribute data showed a progressive relationship from low to high risk level, a sequence from observation to lingering to contact in terms of behavioral nature, and a continuity from normal to suspicious to potential damage in terms of event category. Because the three groups of data had temporal continuity and spatial consistency, they needed to be integrated into a single continuous event through attribute aggregation to avoid misidentifying them as three independent risk points and to make the judgment on vehicle safety status more complete.

[0111] Finally, the processing results obtained from attribute association and attribute aggregation are integrated into a complete safety judgment result, so that each vehicle can obtain a unified and interpretable vehicle safety status, which is then incorporated into the overall parking lot safety management system as the final output.

[0112] In this embodiment, in step S701, risk attribute analysis is performed on each combination of feature elements according to preset risk assessment rules, so that the vehicle status and personnel activities form structured risk attribute data, thereby providing clear and comparable risk basis for subsequent comprehensive judgment; in step S702, attribute association is performed on the risk attribute data of the same combination, and attribute aggregation is performed on the risk attribute data between different combinations, so that the risk attribute data are consistent in expression within the combination and form a continuous and coherent event structure between combinations, thereby realizing a holistic risk understanding of the vehicle status and personnel activities; based on this, in the entire technical solution, through the data-driven expression and event-driven integration of risk attributes, the judgment process of vehicle safety status is improved from single-element identification to comprehensive analysis within and between combinations, thereby achieving a more accurate and interpretable vehicle safety status identification capability.

[0113] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0114] Based on the same inventive concept, this application also provides a vision-based parking monitoring device for implementing the vision-based parking monitoring method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more vision-based parking monitoring device embodiments provided below can be found in the limitations of the vision-based parking monitoring method described above, and will not be repeated here.

[0115] In an exemplary embodiment, as shown in FIG2, a parking lot monitoring device based on vision detection is provided, including: an acquisition module 101, an occupancy analysis module 102, and a security analysis module 103, wherein: the acquisition module 101 is used to acquire the corresponding monitoring video of the parking lot based on the cameras deployed in the parking lot; the occupancy analysis module 102 is used to perform image pixel analysis on the monitoring video based on the image features of the parking spaces to obtain the occupancy status of each parking space in the parking lot, and to provide feedback on the occupancy status of all parking spaces; the security analysis module 103 is used to combine vehicle status features and personnel activity features to perform target detection analysis on the monitoring video to obtain the vehicle safety status of each vehicle in the parking lot, and to issue alarms for abnormal vehicle safety status.

[0116] The modules in the aforementioned vision-based parking monitoring device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0117] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the above embodiments.

[0118] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above embodiments.

[0119] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods.

[0120] 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.

[0121] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A parking lot monitoring method based on visual detection, characterized in that, The method includes: acquiring surveillance video of the parking lot based on cameras deployed in the parking lot; performing image pixel analysis on the surveillance video based on parking space image features to obtain the parking space occupancy status of each parking space in the parking lot, and providing feedback on the occupancy status of all parking spaces; combining vehicle status features and personnel activity features to perform target detection analysis on the surveillance video to obtain the vehicle safety status of each vehicle in the parking lot, and issuing alarms for abnormal vehicle safety status.

2. The method according to claim 1, characterized in that, The step of acquiring surveillance video of the parking lot based on cameras deployed in the parking lot includes: pre-deploying cameras in the entrance / exit area, passage area, and parking space area of ​​the parking lot; determining the imaging parameters of the corresponding cameras according to the scene characteristics of each area to acquire regional surveillance video of the parking lot in each area; classifying and storing the regional surveillance video of each area according to multi-dimensional acquisition attributes to obtain surveillance video represented by a video index structure, which can be used to call the matching regional surveillance video before image pixel analysis and target detection analysis.

3. The method according to claim 1, characterized in that, The step of performing image pixel analysis on the surveillance video based on parking space image features to obtain the parking space occupancy status of each parking space in the parking lot includes: performing edge extraction and edge enhancement on the surveillance video based on the boundary features in the parking space image features to obtain the target image region of each parking space in the surveillance video; and performing image pixel analysis on each target image region in the surveillance video based on the pixel change features in the parking space image features to obtain the parking space occupancy status of each parking space in the parking lot.

4. The method according to claim 3, characterized in that, The step of performing image pixel analysis on each target image region in the monitoring video based on the pixel change features in the parking space image features to obtain the parking space occupancy status of each parking space in the parking lot includes: performing parking scene correlation analysis on the multi-dimensional pixel change pattern of each target image region in the monitoring video based on the pixel change features in the parking space image features to obtain comprehensive change information corresponding to each target image region, which is used as the parking space occupancy status of each parking space in the parking lot.

5. The method according to claim 1, characterized in that, The method of combining vehicle status characteristics and personnel activity characteristics to perform target detection analysis on the surveillance video to obtain the vehicle safety status of each vehicle in the parking lot includes: performing vehicle target-based detection analysis on the surveillance video based on vehicle status characteristics to obtain vehicle detection results for each vehicle during parking; performing personnel target-based detection analysis on the surveillance video based on personnel activity characteristics of each vehicle within its corresponding physical range to obtain personnel detection results for each vehicle within its corresponding physical range; and combining the combination relationship between vehicle detection results and personnel detection results corresponding to the same vehicle to obtain the vehicle safety status of the corresponding vehicle, thereby integrating the results to obtain the overall vehicle safety status of each vehicle in the parking lot.

6. The method according to claim 5, characterized in that, The method of combining the vehicle detection results and personnel detection results corresponding to the same vehicle to obtain the vehicle safety status of the corresponding vehicle includes: obtaining the vehicle feature elements of the vehicle detection results and the personnel feature elements of the personnel detection results from the vehicle detection results and personnel detection results corresponding to the same vehicle; combining each vehicle feature element and each personnel feature element according to the correlation between the feature elements to obtain multiple feature element combinations; and performing joint analysis on the multiple feature element combinations according to the risk attributes reflected by each feature element combination to obtain the vehicle safety status of the corresponding vehicle.

7. The method according to claim 6, characterized in that, The step of jointly analyzing multiple feature element combinations based on the risk attributes reflected by each feature element combination to obtain the vehicle safety status of the corresponding vehicle includes: performing risk attribute analysis on each feature element combination according to the risk assessment rules preset based on the parking scenario to obtain risk attribute data corresponding to each feature element combination; performing attribute association on the risk attribute data corresponding to the same feature element combination; and performing attribute aggregation on the risk attribute data corresponding to different feature element combinations to obtain the vehicle safety status of the corresponding vehicle.

8. A parking lot monitoring device based on vision detection, characterized in that, The device includes: an acquisition module for acquiring surveillance video of the parking lot based on cameras deployed in the parking lot; an occupancy analysis module for performing image pixel analysis on the surveillance video based on parking space image features to obtain the parking space occupancy status of each parking space in the parking lot, and providing feedback on the occupancy status of all parking spaces; and a security analysis module for performing target detection analysis on the surveillance video by combining vehicle status features and personnel activity features to obtain the vehicle security status of each vehicle in the parking lot, and issuing alarms for abnormal vehicle security status.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images