Video recognition-based berth positioning system and method

Abstract

The invention discloses a berth positioning system and method based on video recognition, and relates to the technical field of intelligent transportation, and the system comprises a calibration module which generates berth calibration data based on a berth type, and stores the data to a cloud server; the edge calculation service module is configured to judge the spatial relationship between the vehicle and the berth through a berth type adaptive algorithm based on the real-time latitude and longitude coordinates of the inspection vehicle and the berth calibration data; and the vehicle identification module records license plate information when determining that the vehicle is in the berth, and actively snapshots an empty berth image when no vehicle is in the berth and uploads the empty berth image to the cloud. According to the scheme of the invention, the accuracy, environmental adaptability and long-term operation stability of parking state recognition can be remarkably improved, and a comprehensive solution with high robustness and low misjudgment rate is provided for urban intelligent parking management.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and in particular to a parking space positioning system and method based on video recognition. Background Technology

[0002] With the surge in urban vehicle ownership, efficient management of on-street parking spaces has become crucial for alleviating parking difficulties. Current mainstream parking management technologies mainly fall into two categories: **Geomagnetic + PDA Solution:** This solution uses geomagnetic sensors to detect vehicle presence, combined with manual input of license plate information via handheld PDAs. This method is highly reliant on manual labor, with each parking attendant managing only a limited number of spaces, resulting in high operating costs. **Video Equipment Solution:** This includes high-position video cameras and low-position video monitoring stations, relying on video technology to identify vehicle information within parking spaces. However, its accuracy is limited by environmental factors, and the high cost of constructing a single location limits its coverage to only a portion of parking spaces, making it less economical.

[0003] Furthermore, existing methods generally suffer from the following drawbacks: 1. Traditional GPS calibration of berth latitude and longitude requires extensive manual data collection, and the signal is easily affected by tall buildings and trees, leading to calibration errors; 2. Fixed-point capture schemes are ill-suited for handling irregular behaviors such as parking across berths or at angles, resulting in high rates of missed detections and false detections; 3. The storage method for berth calibration information is singular and lacks anti-tampering mechanisms, easily leading to management disputes. Therefore, based on the above challenges, this invention proposes a berth positioning system and method based on video recognition. Summary of the Invention

[0004] Purpose of the invention To address the aforementioned issues, the present invention aims to provide a berth positioning system and method based on video recognition. This system addresses the problems of low positioning accuracy, poor environmental adaptability, and insufficient data security in traditional berth inspection technologies. Through multi-source data fusion and intelligent algorithm optimization, it achieves high-precision and dynamic berth status identification and management, overcomes the impact of GPS signal interference and berth line obstruction on positioning, reduces missed detection and false detection rates, and improves system response speed and robustness.

[0005] Technical solution To achieve the above objectives, this invention provides a parking space positioning system and method based on video recognition. This scheme generates parking space type, virtual sequence number, and latitude / longitude information based on image segmentation and manual verification. It combines multi-sensor data from GPS, IMU, and video odometer, and optimizes positioning accuracy through Kalman filtering. For perpendicular parking spaces, it calculates the vertical distance threshold between the real-time position of the inspection vehicle and the calibration point, dynamically adjusting the judgment logic. For parallel parking spaces, it optimizes the capture trigger mechanism by considering the geometric relationship between the driving direction and the calibration range. Furthermore, it utilizes an LSTM network to predict abnormal parking space occupancy and aligns vehicle recognition timestamps and location data through a spatiotemporal correlation module to generate a time-series graph to correct misjudgments.

[0006] In a first aspect, the present invention provides a berth positioning system based on video recognition, comprising: The calibration module generates berth calibration data containing berth number, latitude and longitude coordinates, and virtual sequence number based on berth type, and stores the data in a cloud server; The edge computing service module is configured to determine the spatial relationship between the vehicle and the parking space based on the real-time latitude and longitude coordinates of the inspection vehicle and the parking space calibration data, using an adaptive algorithm for parking space type. The vehicle recognition module records the license plate information when it determines that a vehicle is in a parking space, and actively captures images of empty parking spaces when there are no vehicles and uploads them to the cloud. The inspection vehicle is equipped with an edge computing service system, a positioning and data acquisition module, and a vehicle recognition module. The anomaly correction module predicts and corrects misjudgments by analyzing the temporal dependence and spatial distribution characteristics of berth occupancy status. The correction function is as follows:

[0007] In the formula, To adjust the weights; It is the attenuation factor; This represents the cumulative value of historical errors; The slope coefficient; This represents the current confidence level.

[0008] Based on long short-term memory networks and spatiotemporal correlation analysis, this model achieves accurate prediction and correction of abnormal events by mining the temporal dependence and spatial distribution characteristics of berth occupancy status. It can significantly improve the system's adaptability to dynamic scenarios, effectively suppress the propagation of misjudgments, and enhance long-term operational stability.

[0009] Furthermore, the calibration module identifies the direction of the berth line through image segmentation technology and generates berth type labels by combining manual verification. Based on the continuity of the berth calibration points and the driving direction, it dynamically generates a virtual berth sequence that is independent of the physical number.

[0010] Furthermore, the berth calibration data stored on the cloud server is encrypted using blockchain technology to ensure the immutability and traceability of the calibration information.

[0011] Furthermore, the system employs multi-sensor fusion technology to collect the latitude and longitude coordinates of the inspection vehicle in real time, including GPS, inertial navigation unit and video odometer, and optimizes the accuracy of the latitude and longitude coordinates through Kalman filtering algorithm.

[0012] Furthermore, the berth type adaptive algorithm includes: For perpendicular parking spaces, the vertical distance between the real-time position of the inspection vehicle and the parking space calibration point is calculated. If the vertical distance exceeds the preset threshold, the vehicle is determined to be outside the parking space. For parallel parking spaces, the capture trigger logic is dynamically adjusted based on the geometric relationship between the inspection vehicle's driving direction and the parking space's marked range.

[0013] Furthermore, the edge computing service module generates a time-series graph of parking space occupancy status by spatiotemporally aligning the vehicle identification timestamp with the inspection vehicle location data, and predicts parking space occupancy anomalies by using a long short-term memory network trained based on historical data, thereby correcting misjudgment results.

[0014] Furthermore, the vehicle recognition module improves the accuracy of license plate recognition in low-light or occluded scenes by performing feature alignment and fusion on continuously captured vehicle images, and identifies vehicle types based on a deep learning model and associates them with license plate information.

[0015] Furthermore, the system receives cloud commands to switch inspection modes, provides real-time feedback on inspection status to the management platform, and dynamically allocates computing resources based on task priority to reduce the power consumption of the edge computing service system.

[0016] Furthermore, it also includes a threshold adaptive module, used to dynamically adjust the vertical distance threshold based on berth length and environmental complexity. The calculation formula is as follows:

[0017] In the formula, The vertical distance threshold; This is the berth type coefficient; This refers to the berth length. Rate the complexity of the environment.

[0018] By dynamically adjusting the thresholds for determining perpendicular and parallel berths, and combining berth length and environmental complexity factors, the algorithm optimizes the logic for determining the spatial relationship between vehicles and berths. This algorithm significantly improves the accuracy of berth status recognition, enhances the system's adaptability to different berth types and complex environments, and thus reduces the risk of false positives and false negatives.

[0019] In a second aspect, the present invention also provides a berth positioning method based on video recognition, the method being based on the system described in the first aspect above, comprising: Berth calibration data is generated based on berth type, and the data is stored in a cloud server; Based on the real-time latitude and longitude coordinates of the inspection vehicle and the parking space calibration data, the spatial relationship between the vehicle and the parking space is determined through an adaptive algorithm for parking space type. When a vehicle is located in a parking space, the license plate information is recorded, and when there is no vehicle in the parking space, an image of the empty parking space is captured and uploaded to the cloud. By analyzing the temporal dependence and spatial distribution characteristics of berth occupancy status, misjudgments can be predicted and corrected.

[0020] Thirdly, the present invention also provides a computer device, including a management platform and a memory, wherein the management platform is connected to the memory, the memory is used to store computer programs, and the management platform is used to execute the computer programs stored in the memory, so that the computer device performs the aforementioned video recognition-based berth positioning method.

[0021] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a management platform, implements the aforementioned video recognition-based berth positioning method.

[0022] This invention integrates dynamic parking space calibration, multi-sensor fusion positioning, parking space type adaptive threshold algorithm, and a spatiotemporal correlation LSTM anomaly correction model. It processes vehicle identification data and location information in real time through edge computing and uses blockchain to encrypt and store parking space calibration data. The system can adapt to the geometric differences between vertical and parallel parking spaces, optimize the logic for determining the spatial relationship between vehicles and parking spaces, and use deep learning to predict and correct abnormal events. This significantly improves the accuracy, environmental adaptability, and long-term operational stability of parking space status identification. Simultaneously, through encrypted storage and energy consumption optimization mechanisms, it achieves a synergistic improvement in data security and system energy efficiency, providing a highly robust and low-false-rate comprehensive solution for smart parking management in cities.

[0023] Beneficial effects By implementing the video recognition-based berth positioning system and method provided by the present invention, the following technical effects are achieved: (1) This application identifies the direction of berth lines through image segmentation technology, generates berth type labels by combining manual verification, and dynamically generates a virtual berth sequence independent of physical numbers based on the continuity of berth calibration points and driving direction. This solves the problems of low efficiency and easy error in traditional manual calibration, and improves the flexibility of calibration data; the virtual sequence number does not depend on the physical number, adapts to the scenario of dynamic adjustment of berths, and enhances the robustness of data management.

[0024] (2) By integrating multi-source data such as GPS, inertial navigation unit, and video odometer, and optimizing positioning accuracy through Kalman filtering algorithm, the interference of signal obstruction and drift on latitude and longitude calibration is reduced. The reliability of determining the spatial relationship of parking spaces is significantly improved, providing a high-confidence position reference for subsequent vehicle-parking space matching.

[0025] (3) To address the geometric differences between perpendicular and parallel parking spaces, a differentiated judgment logic is designed: for perpendicular parking spaces, the vertical distance threshold between the inspection vehicle and the parking space calibration point is calculated; for parallel parking spaces, the capture trigger mechanism is dynamically adjusted based on the geometric relationship between the driving direction and the calibration range. This optimizes the system's ability to identify non-standard behaviors such as parking across parking spaces and angled parking, reducing the missed detection rate and false detection rate.

[0026] (4) Blockchain technology is used to encrypt and store berth marking data to ensure data integrity and traceability. At the same time, the energy efficiency of the edge computing service system is optimized by dynamically allocating computing resources. This can improve the data security protection level and the system energy efficiency ratio, and achieve synergistic optimization of security and sustainability.

[0027] (5) Based on long short-term memory networks and spatiotemporal correlation analysis, this model achieves accurate prediction and correction of abnormal events by mining the temporal dependence and spatial distribution characteristics of berth occupancy status. It can significantly improve the system's adaptability to dynamic scenarios, effectively suppress the propagation of misjudgments, and enhance long-term operational stability.

[0028] (6) By dynamically adjusting the thresholds for determining perpendicular and parallel berths, and combining berth length and environmental complexity factors, the algorithm optimizes the logic for determining the spatial relationship between vehicles and berths. This algorithm can significantly improve the accuracy of berth status recognition, enhance the system's adaptability to different berth types and complex environments, thereby reducing the risk of misjudgment and missed detection. Attached Figure Description

[0029] To make the above-described video recognition-based berth positioning system and method of the present invention more apparent and understandable, the accompanying drawings used in the specific embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0030] Figure 1 This is a flowchart illustrating a berth positioning method based on video recognition. Figure 2 This diagram shows the layout of the first and last berths. Figure 3 This diagram shows the berth marking points. Figure 4 Diagram showing a vertical berth; Figure 5 This diagram illustrates edge computing judgment. Figure 6 This is a schematic diagram of a parallel berth. Detailed Implementation

[0031] Example 1: A video recognition-based berth positioning system and method are provided. The system includes: a calibration module, which generates berth calibration data containing berth number, latitude and longitude coordinates, and virtual sequential number based on berth type, and stores the data on a cloud server; an edge computing service module, configured to determine the spatial relationship between vehicles and berths based on the real-time latitude and longitude coordinates of an inspection vehicle and the berth calibration data, using a berth type adaptive algorithm; a vehicle recognition module, which records license plate information when a vehicle is located in a berth, and actively captures images of empty berths when no vehicle is present, uploading them to the cloud; and an inspection vehicle, which carries the edge computing service system, the positioning and acquisition module, and the vehicle recognition module. The method flow is as follows: Figure 1 As shown. The details are as follows.

[0032] The video inspection vehicle includes a calibration module and an inspection vehicle device. The calibration module is used to calibrate the latitude and longitude information of the berths to determine which berth a vehicle is located in. The inspection vehicle device is used to identify vehicle information and acquire the latitude and longitude information of the video inspection vehicle in real time.

[0033] The main function of the calibration module is to match berth information with latitude and longitude information. The specific matching process includes: The type of parking space is determined by visual inspection. The parking space types include perpendicular parking spaces and parallel parking spaces. Perpendicular parking spaces are those where the parking direction is perpendicular to the vehicle's driving direction; parallel parking spaces are those where the parking direction is parallel to the vehicle's driving direction. Parking space types must be consistent within the same road segment. For example, in a scenario where a road segment has both perpendicular and parallel parking spaces, two road segments need to be created during the parking space marking process, and then the parking spaces are marked in the corresponding road segments according to their parking space types.

[0034] The basic information of a berth includes the berth number, berth type, and first and last berth. The first and last berths are recorded visually to indicate the start and end of consecutive berths within a road segment, such as... Figure 2 As shown, the inspection vehicle device determines whether to switch the identification method based on the first and last parking space information.

[0035] like Figure 3 As shown, the latitude and longitude of each berth are determined based on two calibration points on the side closest to the road in the direction of travel. For consecutive berths, the last calibration point of the previous berth is used as the starting calibration point for the current berth. Since berth numbers do not reflect the sequential relationship between berths, virtual berth numbers are generated simultaneously during berth calibration to determine the sequential order of berths. After being collected, berth information is recorded on the inspection vehicle's device and in the cloud server.

[0036] The patrol vehicle device consists of an edge computing service module, a positioning and acquisition module, and a vehicle identification module. The patrol vehicle device identifies vehicle information through the vehicle identification module and acquires latitude and longitude data in real time at a certain frequency through the positioning and acquisition module. Finally, the data is handed over to the edge computing service module for calculation.

[0037] The vehicle recognition module can identify vehicle information including license plate number, license plate color, and vehicle type.

[0038] When the inspection vehicle device is operating normally, the positioning and acquisition module acquires the device's location information, i.e., latitude and longitude coordinates, at a certain frequency in real time. The vehicle recognition module identifies the captured vehicle and records the latitude and longitude coordinates acquired by the positioning and acquisition module at that moment. This information is then passed to the edge computing service module for calculation and judgment. By comparing the latitude and longitude coordinates of the inspection vehicle device when the vehicle is captured with the coordinates of the designated berth, the system determines whether the vehicle is located within the berth. If the vehicle is located within the berth, it determines which berth it is in and uploads the result to the cloud server for storage and recording.

[0039] After passing a parking space, the inspection vehicle device uses the vehicle recognition module and edge computing service module to determine whether there is a vehicle parked in the parking space it is currently passing, and uploads a captured image to the cloud server to verify the true status of the parking space.

[0040] The inspection vehicle's equipment uses different judgment methods depending on the type of berth.

[0041] Vertical berths, such as Figure 4 As shown, the patrol vehicle's travel direction is perpendicular to the vehicle's parking direction. At this point, the patrol vehicle's latitude and longitude coordinates when taking the picture are between the two points of the parking space. If it is on the horizontal coordinate axis, vehicles outside this area are considered not to belong to that parking space. If it is on the vertical coordinate axis, with the patrol vehicle's latitude and longitude coordinates at the time of the picture as the foot of the perpendicular, a perpendicular line is drawn to the straight line formed by the coordinates of the two points of the parking space. The distance from the foot of the perpendicular to the straight line is the distance between the patrol vehicle and the parking space. If this distance exceeds a certain range or is less than a certain range, the vehicle is considered not to belong to that parking space.

[0042] Edge computing judgment such as Figure 5 As shown, if the latitude and longitude of the inspection vehicle's device are not between the berth markers, and the vehicle recognition system still fails to capture a vehicle after the inspection vehicle has passed the berth markers in sequence, it is determined that the berth is empty, and the vehicle recognition system is activated to capture an image and upload it to the cloud server for recording.

[0043] like Figure 6As shown, in parallel parking spaces, the inspection vehicle's driving direction is parallel to the vehicle's parking direction. The vehicle recognition module can capture images of vehicles parked in the parking space even when it is outside the two marked points. Vehicles within the two marked points are considered not to be in the current parking space. If the inspection vehicle moves away from the two marked points and the vehicle recognition module still cannot recognize a vehicle, the parking space is considered empty, and the vehicle recognition module is automatically activated to capture an image as evidence and upload it to the cloud server.

[0044] Example 2: Building upon the aforementioned embodiments, and considering that traditional anomaly correction relies on static rules and struggles to handle nonlinear correlations in time-series data, a spatiotemporal correlation-LSTM anomaly correction model is added. This model combines a spatiotemporal correlation module with an LSTM network to predict and correct misjudgments by analyzing the temporal dependence and spatial distribution characteristics of berth occupancy status, thereby improving system robustness.

[0045] Define the correction function as a function of historical error and current confidence level:

[0046] In the formula, To adjust the weights, the extent to which historical errors affect the current adjustment is determined; This is the attenuation factor, used to control the attenuation rate of historical errors; This is the cumulative value of historical errors, used to reflect the systematic bias of the model in historical data; This is the slope coefficient, used to adjust the sensitivity of the current confidence level to the correction amount; This represents the current confidence level, used to dynamically suppress the correction magnitude under low confidence conditions, preventing interference from noise or outliers.

[0047] The input to the LSTM network includes time-series data of berth occupancy status over the past 5 minutes and environmental parameters, and the output includes the probability of anomalies in the next 2 minutes; if the probability of anomalies is high, the edge computing service module is triggered to re-verify the data.

[0048] Suppose a vertical berth experiences the following state changes over 5 consecutive minutes, with environmental parameters fluctuating in real time. The system uses an LSTM network to predict the probability of anomalies in the next 2 minutes and triggers correction logic.

[0049] The input parameters and calculation results are shown in Table 1.

[0050] Table 1. Input parameters and calculation results

[0051]

[0052] The LSTM network predicts an anomaly probability of 0.76 (>0.7) for the next 2 minutes, triggering a re-verification.

[0053] The results of the spatiotemporal correlation-LSTM anomaly correction model are shown in Table 2.

[0054] Table 2. Summary of the effects of the spatiotemporal correlation-LSTM anomaly correction model

[0055] According to the experimental data, LSTM correction reduced the false positive rate of abnormal events from 3.1% to 0.4%; the correction decision delay was shortened from 1.2 seconds to 0.3 seconds; and after 24 hours of continuous operation, the system accuracy fluctuation range decreased from ±2.5% to ±0.7%. The results show that this model, by integrating time series analysis and deep learning prediction capabilities, significantly improves the system's robustness to dynamic abnormal events. When berth occupancy status changes frequently or environmental parameters fluctuate drastically, the model can accurately capture the spatiotemporal correlation of abnormal events and suppress the propagation of misjudgments. Furthermore, this technique effectively optimizes the system's ability to correct accumulated errors over long-term operation, ensuring high stability and continuity of inspection tasks.

[0056] Example 3: Building upon the aforementioned embodiments, and considering the differences in geometric characteristics between perpendicular and parallel berths, a dynamic adaptive threshold algorithm for berth type is added to address the inability of traditional fixed threshold determination methods to adapt to changes in berth length and complex environmental interference. By introducing a berth length ratio coefficient and an environmental complexity weight, the determination threshold is adjusted in real time, achieving precise adaptation between berth type and scenario.

[0057] The vertical distance threshold is dynamically calculated based on the berth length and environmental complexity, using the following formula:

[0058] In the formula, The vertical distance threshold; This is the berth type coefficient; This refers to the berth length. Score the environmental complexity. Data collected in real time by the inspection vehicle, including obstruction density, light intensity, rain / fog level, and GPS signal strength, are used to calculate data in real time via a convolutional neural network. The calculation formula is as follows: In the formula, The raw image data captured in real time by the inspection vehicle's camera. This refers to the set of real-time readings from non-video sensors mounted on the inspection vehicle.

[0059] Berth type coefficient Reflecting the differences between perpendicular and parallel berths in terms of geometric characteristics and inspection perspective: Vertical berth ( ): The vehicle is perpendicular to the direction of travel of the inspection vehicle, and a smaller threshold is required to accurately capture lateral positional deviation; Parallel berths ( ): When vehicles and inspection vehicles are parked in the same direction, a larger threshold is required to accommodate longitudinal position fluctuations.

[0060] If the vertical distance of the inspection vehicle If the vehicle is located within the parking space, it is determined that the vehicle is in the parking space; otherwise, it is marked as an abnormal parking space and a secondary verification is triggered.

[0061] Verification shows that, while achieving an average error similar to the aforementioned embodiments, the dynamic threshold reduces the false positive rate for perpendicular berths from 6.2% to 0.8%, and the false negative rate for parallel berths from 7.5% to 1.3%; even in occluded scenarios, the accuracy remains at 98.5%. The results demonstrate that this algorithm significantly improves the accuracy of berth status recognition by dynamically adjusting the judgment thresholds for perpendicular and parallel berths. In complex environments, the system can adaptively adjust the judgment logic, effectively reducing the risk of false positives and false negatives caused by environmental interference. Furthermore, the algorithm enhances compatibility with different berth types, ensuring that the inspection vehicle can stably output high-confidence recognition results in berth scenarios with varying geometric features.

[0062] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable non-transitory storage media containing computer-usable program code.

[0063] The present invention can provide computer program instructions to a management platform of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing equipment to produce a machine, such that the instructions executed by the management platform of the computer or other programmable data processing equipment produce means for implementing the system.

[0064] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that perform the functions of the system.

[0065] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions of the system.

Claims

1. A berth positioning system based on video recognition, characterized in that, include: The calibration module generates berth calibration data based on the berth type and stores the data on a cloud server. The edge computing service module is configured to determine the spatial relationship between the vehicle and the parking space based on the real-time latitude and longitude coordinates of the inspection vehicle and the parking space calibration data, using an adaptive algorithm for parking space type. The vehicle recognition module records the license plate information when it determines that a vehicle is in a parking space, and actively captures images of empty parking spaces when there are no vehicles and uploads them to the cloud. The anomaly correction module predicts and corrects misjudgments by analyzing the temporal dependence and spatial distribution characteristics of berth occupancy status. The correction function is as follows: In the formula, To adjust the weights; It is the attenuation factor; This represents the cumulative value of historical errors; The slope coefficient; This represents the current confidence level.

2. The system according to claim 1, characterized in that: The calibration module identifies the direction of the berth line through image segmentation technology and generates berth type labels by combining manual verification. Based on the continuity of the berth calibration points and the driving direction, it dynamically generates a virtual berth sequence that is independent of the physical number.

3. The system according to claim 1, characterized in that: The system uses multi-sensor fusion technology to collect the latitude and longitude coordinates of the inspection vehicle in real time, and optimizes the accuracy of the latitude and longitude coordinates through Kalman filtering algorithm.

4. The system according to claim 1, characterized in that: The berth type adaptive algorithm includes: For perpendicular parking spaces, the vertical distance between the real-time position of the inspection vehicle and the parking space calibration point is calculated. If the vertical distance exceeds the preset threshold, the vehicle is determined to be outside the parking space. For parallel parking spaces, the capture trigger logic is dynamically adjusted based on the geometric relationship between the inspection vehicle's driving direction and the parking space's marked range.

5. The system according to claim 1, characterized in that: The edge computing service module generates a time-series map of parking space occupancy status by spatiotemporally aligning vehicle identification timestamps with inspection vehicle location data, and predicts parking space occupancy anomalies by using a long short-term memory network trained based on historical data.

6. The system according to claim 1, characterized in that: The vehicle recognition module reduces environmental interference by aligning and fusing features of continuously captured vehicle images, and identifies vehicle types based on a deep learning model, which is then linked to license plate information.

7. The system according to claim 1, characterized in that: The system receives cloud commands to switch inspection modes, provides real-time feedback on inspection status to the management platform, and dynamically allocates computing resources based on task priority.

8. The system according to any one of claims 1-7, characterized in that: It also includes a threshold adaptive module, used to dynamically adjust the vertical distance threshold based on berth length and environmental complexity. The calculation formula is as follows: In the formula, The vertical distance threshold; This is the berth type coefficient; This refers to the berth length. Rate the complexity of the environment.

9. A berth positioning method based on video recognition, characterized in that: The method is implemented based on the system described in any one of claims 1-8: The method includes: Berth calibration data is generated based on berth type, and the data is stored in a cloud server; Based on the real-time latitude and longitude coordinates of the inspection vehicle and the parking space calibration data, the spatial relationship between the vehicle and the parking space is determined through an adaptive algorithm for parking space type. When a vehicle is located in a parking space, the license plate information is recorded, and when there is no vehicle in the parking space, an image of the empty parking space is captured and uploaded to the cloud. By analyzing the temporal dependence and spatial distribution characteristics of berth occupancy status, misjudgments can be predicted and corrected.

10. A computer-readable storage medium storing a computer program, characterized in that: The computer program is executed by the processor to perform the method of claim 9.

Images