Large-dip-angle license plate recognition and berth management equipment

Abstract

The invention discloses large-dip-angle license plate recognition and berth management equipment. The equipment is composed of four functional modules including a license plate recognition module, a berth management module, a data processing module and an environment adaptation module. The license plate recognition module adopts a wide-angle lens and a high-resolution camera to support large-range shooting of vehicle photos. The berth management module monitors the occupation state of berths in real time through a geomagnetic sensor and a camera. And the data processing module integrates the license plate recognition result and the berth state data. And the environment adaptation module performs corresponding parameter adjustment according to different deployment environments. The intelligent and high-efficiency parking lot management system is created by integrating multiple innovative technologies such as large-dip-angle license plate recognition, flow-saving parking space detection and routing inspection, full-process optimization from vehicle entering recognition and parking space accurate management to equipment operation and maintenance guarantee is achieved, the parking lot operation management level and user experience are greatly improved, and the parking lot management efficiency is improved. Wide application prospects and popularization values are realized.

Description

Technical Field The present invention belongs to the technical field of intelligent transportation management, and particularly relates to a large inclination angle license plate recognition and berth management device, which is applicable to scenarios such as parking lots and roadside berths, and can efficiently recognize vehicle information of license plates with large inclination angles and realize intelligent berth management. Background Art

[0001] There are many challenges in the existing license plate recognition and berth management technologies. In terms of license plate recognition, traditional devices are sensitive to the angle of the license plate. When the parking angle of the vehicle is large (such as diagonal parking or tilted front), the recognition accuracy will drop significantly. In addition, factors such as bad weather, complex lighting conditions, and background interference will also seriously affect the accuracy of license plate recognition. In terms of berth management, the existing systems generally rely on manual or simple sensors and cannot accurately grasp the berth status in real time, resulting in low management efficiency. More importantly, the license plate recognition system and the berth management system often act independently, and the data cannot be effectively integrated, restricting the realization of intelligent management. Summary of the Invention

[0002] To solve the above problems, the present invention provides a large inclination angle license plate recognition and berth management device, which realizes efficient license plate recognition and berth management by combining high-precision image processing technology and intelligent berth management algorithms.

[0003] The present invention adopts the following technical solutions: The device is composed of four functional modules: a license plate recognition module, a berth management module, a data processing module, and an environment adaptation module. The license plate recognition module uses a wide-angle lens and a high-resolution camera to support large-range shooting of vehicle photos. The license plate recognition module performs image preprocessing (such as denoising, enhancement, and skew correction) on the collected images to improve the image quality; secondly, it accurately locates the license plate area through a target detection algorithm, and performs geometric correction on the license plate with a large inclination angle; by calculating the inclination angle of the license plate, it dynamically adjusts the recognition algorithm, and combines the license plate recognition model to recognize the corrected license plate characters. The berth management module real-time monitors the occupancy status of the berths through geomagnetic sensors and cameras; according to the berth occupancy situation and vehicle recognition information, it dynamically allocates berths to optimize the utilization rate of parking resources. The data processing module integrates the license plate recognition results and the berth status data, and uploads the integrated information to the server database through wireless communication (such as 4G\5G); at the same time, it accepts remote control from the server to update algorithm parameters and perform fault diagnosis. The environment adaptation module makes corresponding parameter adjustments for different deployment environments. For example, in low-light conditions, it automatically activates the fill light or infrared camera to ensure clear images; for outdoor deployment, such as in bad weather like rain, snow, and haze, it optimizes the image processing algorithm to improve the recognition accuracy.

[0004] Step S1: After the device is powered on, sensor initialization and self-check are performed first. Hardware modules such as the geomagnetic sensor and camera are started, and the device checks whether each module is working properly. If the self-check fails, the device enters the fault mode and reports to the server; if the self-check passes, the device enters the low-power mode, the geomagnetic sensor detects the berth status at a low frequency, and the camera and 4G module are turned off to save power.

[0005] Step S2: In the low-power mode, the geomagnetic sensor detects the berth status at a low frequency (such as once per second). When the geomagnetic sensor detects a magnetic field change, the device determines whether a vehicle enters or leaves the berth. If a vehicle entry is detected, the device triggers the vehicle detection logic; if a vehicle departure is detected, the device executes the departure detection logic.

[0006] Step S3: If a vehicle is detected entering the berth, the device starts the camera to take a photo and preprocesses the image (including denoising, enhancement, and skew correction). The license plate area is accurately located using object detection algorithms (such as YOLO, SSD, etc.). License plate characters are recognized through the license plate recognition algorithm. After recognition, the device uploads the license plate recognition result, the image, and the parking space status (occupied) to the server through the 4G module. After the upload is completed, the device starts timing and detects whether the vehicle has parked for 10 minutes.

[0007] Step S4: If the vehicle has parked in the berth for 10 minutes, the device starts the camera to take a photo again and uploads the data to ensure the real-time update of vehicle information. After the upload is completed, the device enters the low-power mode and continuously detects whether the vehicle has left.

[0008] Step S5: In the low-power mode, the geomagnetic sensor continuously detects whether the vehicle has left. If a vehicle departure is detected, the device starts the camera to take a photo and uploads the parking space status (unoccupied). After the upload is completed, the device returns to the idle detection state and continues to detect whether a vehicle enters the berth.

[0009] Furthermore, when the device completes the photo-taking and uploading task, it enters the low-power mode. The camera and 4G module are turned off, the geomagnetic sensor operates at a low frequency, and the device power consumption is reduced to the lowest. When the geomagnetic sensor detects a vehicle entering or leaving, the device wakes up and executes the corresponding logic.

[0010] Preferably, if the 4G module fails to upload data, the device temporarily stores the data locally and re-uploads it after the communication is restored; if the geomagnetic sensor or camera fails, the device reports to the server and enters the fault mode; if the device power is insufficient, the device reports to the server and enters the sleep mode.

[0011] Preferably, for the large inclination angle license plate recognition and berth management device of the present invention, the image recognition threshold, license plate feature extraction parameters, etc. used in its internal algorithm can be adaptively adjusted according to different installation angles, lighting conditions, and license plate types. This adaptive adjustment mechanism ensures that the device can still maintain the accuracy of license plate recognition and the effectiveness of berth management in complex and dynamic environments, such as large inclination angle installation, night, or strong light irradiation conditions.

[0012] Furthermore, the present invention flexibly adjusts various parameters in the license plate recognition algorithm, such as brightness adjustment in the image preprocessing stage, contrast enhancement, and the sensitivity of the license plate positioning algorithm, to meet the recognition requirements of different types of license plates (such as ordinary license plates, new energy vehicle license plates, temporary license plates, etc.). At the same time, the berth management module also has a high degree of versatility and can be configured according to the specific requirements of different parking lots, road monitoring, and other scenarios, such as berth occupancy status detection, vehicle stay time recording, etc. In specific implementation, these parameters can be customized according to the on-site environment, license plate features, and management requirements to ensure the stability and efficiency of the entire system.

[0013] In response to the demand for traffic-saving berth detection and management, the present invention introduces an intelligent data transmission strategy. This strategy significantly reduces the unnecessary data transmission volume by intelligently screening key recognition results and events, such as successful license plate recognition, berth status changes, etc., thereby achieving the goal of traffic saving. At the same time, an efficient image compression algorithm and a data processing module are integrated inside the device to further reduce the bandwidth requirements for data transmission.

[0014] The device of the present invention realizes the efficient coordination of license plate recognition and berth management through an integrated system design. The license plate recognition module can accurately handle scenarios such as large inclination angle license plates, complex lighting conditions, and bad weather, and achieve efficient license plate recognition. The berth management module combines the license plate recognition data and real-time berth status monitoring, and dynamically adjusts the berth allocation strategy, which not only improves the utilization rate of parking resources but also effectively reduces the time for vehicle owners to search for parking spaces and traffic congestion in the parking lot. The data processing module plays a key role in the present invention. By integrating the license plate recognition results and berth management data, it forms a unified data interface and uses wireless communication technology to achieve remote data transmission, providing real-time and comprehensive parking space status and vehicle information for parking lot managers. At the same time, the environment adaptation module enables the device to adapt to various complex environments, for example, by dynamically adjusting the camera parameters and enabling auxiliary light sources to ensure the accuracy of recognition in different deployment environments.

[0015] The key problems solved by the present invention include: how to achieve accurate recognition of large - inclination license plates in complex and dynamic environments; how to flexibly configure according to different types of license plates and management requirements; and how to achieve traffic - saving data transmission and efficient patrol while ensuring recognition accuracy. By introducing innovative points such as an adaptive adjustment mechanism, a general - purpose design, and an intelligent data - transmission strategy, the present invention not only improves the accuracy and efficiency of license - plate recognition and berth management, but also reduces the data - transmission cost and meets the requirements of various application scenarios. At the same time, its high configurability and intelligent design also facilitate future expansion and upgrade. Description of the Drawings

[0016] Figure 1 It is the working flowchart of the device in the embodiment;

[0017] Figure 2 It is the system - module diagram of the device in the embodiment. Detailed Implementation Manner

[0018] The technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0019] The indoor parking lot is a representative parking environment. This embodiment takes the large - inclination license - plate recognition and berth - management device in the indoor parking lot as an example for illustration.

[0020] A large - inclination license - plate recognition and berth - management device, combined with traffic - saving berth detection, is used for efficient management of parking in a parking lot. This method effectively improves the utilization rate and management efficiency of parking spaces in the parking lot through a multi - module collaborative - processing mechanism, such as Figure 1 - Figure 2 shown, including the following steps: Step S1: Use a high-definition camera with the ability to shoot at large angles to cover multiple berth angles. Regularly obtain real-time berth images at a preset shooting cycle (every 5 seconds) to ensure that the images are complete and unobstructed. Based on deep learning object detection algorithms, such as YOLOv5, a large number of image datasets labeled with vehicle information are used for training. The training dataset includes various parking lot scenarios, different vehicle types, and different environmental changes (such as lighting, shadows, angles, etc.), enabling the model to have strong generalization ability. In YOLOv5, the image will first pass through the convolutional layer for feature extraction, and the model will automatically identify basic features in the image, such as edges, shapes, textures, etc. Subsequently, YOLOv5 divides the image into multiple grid cells, and each grid is responsible for predicting the targets within its area. Each grid will output several pieces of information, including the category of the target (such as "car") and the coordinates of the bounding box of the target. Each detected target vehicle will have a bounding box, which represents the position of the target in the image. YOLOv5 will regress the parameters of the bounding box of the target, including: the center point coordinates (x, y), width (w), and height (h). The position of the bounding box can help accurately calibrate the vehicle position in the subsequent steps. Use this YOLOv5 model to detect vehicles in the collected images, identify vehicle targets in the images and generate bounding boxes, and at the same time label the vehicle positions. Combine the berth area information of the parking lot to preliminarily judge the berth status, mark suspected idle or occupied berths, reduce misjudgments caused by environmental factors, and provide basic data support for subsequent license plate recognition and precise berth detection. Specifically, the triggering mechanism for the camera to work can be: when the geomagnetic sensor determines that the parking space is occupied, the signal will be transmitted to the management system through the wireless communication module, and the system will then start the camera to take pictures. After the geomagnetic sensor confirms that the vehicle is parked, the camera automatically adjusts the focus to ensure that the license plate can be clearly captured. Based on the ability of large-angle shooting, the camera may need to perform perspective transformation (according to the mapping relationship between the calibration coordinates on the parking lot ground and the image pixel coordinates) to correct the license plate angle in the image and restore it to a horizontal state. Step S2: Perform license plate localization and recognition preprocessing on the vehicle area image detected in Step S1. First, convert the image to a grayscale image to reduce the data processing volume, and then apply edge detection algorithms (such as Sobel or Canny operators) to extract the edge features of the image, and use morphological operations (erosion, dilation) to remove noise and connect broken edges to enhance the license plate contour. For large-angle license plates, use the perspective transformation algorithm to correct the license plate tilt angle according to the mapping relationship between the calibration coordinates on the parking lot ground and the image pixel coordinates to restore the license plate to a horizontal state; then perform character segmentation on the corrected license plate area, and segment out individual character images based on character spacing, width, and connected domain characteristics to prepare data for the subsequent character recognition stage. Specifically: The perspective transformation algorithm is: The position of the license plate in three-dimensional space is (X, Y, Z), and the image coordinates (x, y) are obtained through perspective transformation. The expression of the perspective transformation matrix H is: where H is the perspective transformation matrix; The character segmentation is specifically based on the Mask R-CNN or U-Net segmentation network of deep learning to segment the characters, and the region R of each character is obtained c , and the goal of the model is to optimize the segmentation accuracy by minimizing the intersection over union (IoU) loss: where, is the intersection over union between the true character region and the predicted region, is the character region predicted by the model. Step S3: License plate character recognition and traffic optimization. The segmented character image is input into a deep learning character recognition model (such as a convolutional neural network CNN). The model is trained with a large number of license plate samples and can accurately recognize the character content and output the license plate number; during this process, a traffic-saving strategy is introduced. The model adopts a lightweight architecture design to reduce the occupation of computing resources. At the same time, incremental learning technology is used to update the model parameters in real time according to the new license plate samples collected at the parking lot site to improve the recognition accuracy of special license plates (such as damaged and deformed ones); after recognition, the license plate number is compared with the vehicle information in the parking lot database to determine whether the vehicle is a reserved vehicle, a monthly-parking vehicle, or a temporary parking vehicle, and it is classified and managed, and information such as the vehicle entry time and berth location is recorded. Specifically: Character recognition uses a convolutional neural network or a recurrent neural network to recognize the segmented characters. Let the segmented character image be I(x, y). The character recognition model is trained through deep learning and outputs the character recognition result C i , and the loss function for training is: where y is the true character label, and p i is the character probability predicted by the model. N is the number of characters. The smaller the loss value, the closer the prediction result of the model is to the true label. Minimizing the loss function: By continuously optimizing the loss function, that is, minimizing the loss, the parameters of the model (such as convolutional kernels, fully connected layers, etc.) will be gradually adjusted so that the predicted probability p i output by the model is closer to the true label y. This process usually uses the gradient descent method or its variants (such as Adam) to carry out; For each character, the training process minimizes the cross-entropy loss by continuously adjusting the parameters of the model, making the probability distribution predicted by the model tend to the distribution of the true label. In the recognition stage, when an image is input, the model outputs the predicted probability distribution of each character. According to the principle of maximum probability, the category with the highest probability is selected as the recognition result of the character. For example, the model generates probabilities for 10 categories for each character (assuming there are 10 numeric categories 0 - 9). These probability values represent the confidence level of the model for the character belonging to each category. For a character, the model predicts the following probabilities: p = [0.05, 0.10, 0.15, 0.60, 0.05, 0.02, 0.02, 0.01, 0.01, 0.00] The model will select the character category corresponding to the maximum probability, that is, category 3. Therefore, the recognition result of this character is "3"; Finally, the recognition result is output: By combining the recognition results of each character, the entire license plate number is obtained. For example, if the license plate is "AB1234", the model recognizes each character ("A", "B", "1", "2", "3", "4") in sequence and finally outputs the complete license plate number. Step S4: Precise berth inspection and anomaly judgment. The berth detection and inspection equipment with reduced data flow starts the precise inspection process. Using high-precision sensors installed at key positions of the berth (such as the geomagnetic sensor with the resolution improved to 0.01 gauss), the suspected idle berths are re-detected to accurately judge whether the berth is truly idle, and at the same time, whether the vehicle parking posture is standard (such as whether it presses the line or straddles the parking space); if abnormal situations are detected, such as misjudgment of the berth status or illegal vehicle parking, an alarm message is immediately triggered, and the alarm signal is transmitted to the management center through the wireless communication module. The management center highlights the position of the abnormal berth on the monitoring large screen and pushes a processing task to the inspection terminal. Specifically, whether the vehicle parking posture is standard is calculated through the posture standard detection formula: Assume that the parking space area is a rectangular area, and the position of the license plate area is [x1, y1, x2, y2], corresponding to its upper left and lower right coordinates. Then, whether the vehicle presses the line can be judged by the following inequality: Over = max(0, min(x2, x3) - max(x1, x4)) × max(0, min(y2, y3) - max(y1, y4)) [x3, y3, x4, y4] are the boundary coordinates of the berth area. If the overlapping part is less than a certain preset threshold (such as 10% of the berth area), it is judged that the vehicle does not press the line. In addition, assume that the normal state data value of the berth is S norm (i), if a value significantly different from the normal state is found in the sensor data, the anomaly degree A can be defined i(t), represents the abnormality degree of berth i at time t, and the abnormality degree can be calculated by the following formula: A i (t)=||D i (t)-S norm (i)|| / σ(D i ) Among them, σ(D i ) is the standard deviation of the sensor data at berth i, D i (t) represents the sensor data value (the data value collected by the magnetic field sensor (such as the geomagnetic sensor) can be selected. The sensor is very sensitive to the metal part of the vehicle and can provide high-precision detection data. When the vehicle enters or leaves the parking space, the metal part of the vehicle will affect the magnetic field and cause the sensor signal to change). If A i (t) If it exceeds a certain threshold, it is considered that the berth is abnormal. Step S5: Data fusion and parking space utilization analysis. The license plate recognition data is integrated with the parking inspection data to establish a real-time parking space usage database. Based on this database, the parking space utilization rates in different time periods and different areas are analyzed, and a parking space utilization trend chart is drawn to provide a decision-making basis for parking lot operators, such as reasonably adjusting charging strategies and optimizing the layout of parking space guidance signs according to peak and valley periods; at the same time, data analysis is used to explore potential needs, predict parking space demand in future time periods, make preparations in advance, and improve the overall operation efficiency of the parking lot. Step S6: Equipment self-diagnosis and remote upgrade. The system regularly (such as during the low-peak period of 2-4 a.m. every day) conducts self-diagnosis on the high-angle license plate recognition equipment, flow-saving berth detection and inspection equipment, checks key indicators such as camera image acquisition quality, sensor working status, communication module signal strength, etc. If an abnormality is found, it automatically tries to restart the relevant equipment or module; if the problem persists, a detailed fault report is generated and sent to the operation and maintenance center via remote communication; it has a remote upgrade function. When there is a new algorithm optimization or function improvement, the equipment can automatically receive and update the software version during idle time to ensure that the system is always in the best operating state.

[0021] In this embodiment, the high-definition camera used for image acquisition is connected to the image processing system via 4G communication, and the image data is transmitted according to a preset communication protocol. When the acquisition is started, the camera captures the picture at a set frame rate and transmits the detected image to the back end for processing.

[0022] For step S1: First, ensure that the camera is correctly installed in a suitable position in the parking lot to ensure that the license plate information of each berth and the incoming and outgoing vehicles can be clearly captured. Adjust the shooting angle to meet the large inclination requirement. Connect the camera to the data receiving interface of the image processing system to complete the initialization preparation work. Once started, the system will receive the image data collected by the camera. At the same time, start the traffic-saving berth detection module. Based on the low-power sensor network, this module initially determines whether a berth is idle. The sensor quickly marks the suspected idle berths by sensing the changes in the magnetic field, pressure, or infrared signal above the berth, and transmits this information to the central processing unit.

[0023] For step S2, the parameters of the image recognition algorithm are optimized and adjusted according to the actual lighting conditions and license plate style characteristics of the parking lot. For example, in the area directly irradiated by strong light, enhance the image contrast; for license plate fonts and colors with local characteristics, specifically expand the training samples. In the calibration process of the perspective transformation correction algorithm, with the help of high-precision measurement tools, accurately measure the ground coordinate points of the parking lot to construct an accurate mapping model to ensure the correction effect of large-inclination license plates and improve the subsequent recognition accuracy. The character segmentation algorithm combines with the deep learning model to automatically learn the character segmentation rules in different scenarios to meet the license plate recognition requirements in complex environments.

[0024] If the license plate positioning fails or the recognition accuracy is lower than the preset threshold (such as 85%), automatically mark this image frame as abnormal, generate and output a prompt message of "License plate recognition is abnormal, manual verification of the image is required", and at the same time pause the current processing flow, waiting for manual intervention or re-acquisition of the image; if the license plate recognition is successful and the accuracy meets the standard, then continue the subsequent data processing flow and store the valid data in the database.

[0025] For step S3, the training of the deep learning character recognition model is carried out in cooperation between the cloud and the local. Use the powerful computing resources of the cloud for large-scale initial training, and use the newly collected samples on the local device for incremental fine-tuning, which not only ensures the generalization ability of the model but also can quickly adapt to the changes on the parking lot site; the lightweight architecture design reduces the number of model parameters and the computational complexity through techniques such as model pruning and quantization; during the incremental learning process, set a sample screening mechanism to select representative new samples to update the model to avoid overfitting and continuously improve the recognition performance.

[0026] For step S4, the high-precision sensor is finely calibrated during installation to ensure the accuracy and reliability of the detection data; the wireless communication module uses adaptive signal strength adjustment technology to dynamically optimize the transmission power according to the signal interference situation in the parking lot environment to ensure the timely and stable transmission of alarm information; after receiving the task, the handheld terminal of the patrol personnel has a navigation function and can quickly locate the abnormal berth to improve the processing efficiency.

[0027] For step S5, during the data fusion process, a strict data cleaning and verification mechanism is established to eliminate duplicate and incorrect data, ensuring the accuracy of the parking space usage database. For the analysis of parking space utilization rate, various data analysis methods such as time series analysis and clustering analysis are adopted to deeply explore the data value and provide scientific support for operation decision-making. The trend chart is updated in real time to intuitively display the dynamic changes in parking space usage, assisting managers in quickly understanding the operation status of the parking lot.

[0028] For step S6, during the self-diagnosis process, each device and module outputs a status report according to the preset standard, and the diagnostic program quickly locates the root cause of the problem based on the fault code table. For remote upgrade, technologies such as breakpoint resume and version rollback are adopted to ensure the safety and stability of the upgrade process and avoid system paralysis caused by upgrade failure. After receiving the fault report, the operation and maintenance center can remotely control the device to conduct partial fault troubleshooting and repair, reducing the on-site operation and maintenance costs.

[0029] This embodiment integrates a number of innovative technologies such as large-angle license plate recognition, traffic-saving berth detection and inspection, etc., to create an intelligent and high-performance parking lot management system, realizing the full-process optimization from vehicle entry recognition, accurate berth management to equipment operation and maintenance guarantee, greatly improving the operation and management level of the parking lot and the user experience, and having broad application prospects and promotion value.

[0030] Although this embodiment focuses on the parking lot scenario, the relevant technical principles and architectures can be extended to large vehicle parking and management places such as logistics parks, ports and terminals through adaptive adjustment, and customized development can be carried out to meet the special needs of different fields, further expanding the application boundary of the technology.

Claims

1. A license plate recognition and berth management device with a large inclination angle, characterized in that: It consists of four functional modules: license plate recognition module, parking space management module, data processing module and environment adaptation module; The license plate recognition module uses a wide-angle lens and a high-resolution camera to support a wide range of vehicle photos. The license plate recognition module performs image preprocessing on the collected images to improve image quality. Secondly, the target detection algorithm is used to accurately locate the license plate area and perform geometric correction on license plates with large inclination angles. The recognition algorithm is dynamically adjusted by calculating the inclination angle of the license plate, and the license plate recognition model is combined to recognize the corrected license plate characters. The parking space management module monitors the parking space occupancy status in real time through geomagnetic sensors and cameras; dynamically allocates parking spaces based on parking space occupancy and vehicle identification information to optimize parking resource utilization; The data processing module integrates the license plate recognition results with the parking space status data, and uploads the integrated information to the server database via wireless communication. At the same time, it accepts remote control from the server to update algorithm parameters and fault diagnosis. The environmental adaptation module makes corresponding parameter adjustments for different deployment environments. For example, in low-light conditions, it automatically enables fill lights or infrared cameras to ensure clear images. For outdoor deployments, such as rain, snow, fog and haze, it optimizes the image processing algorithm to improve recognition accuracy.

2. A method applied to the large-inclination license plate recognition and berth management device described in claim 1, characterized in that: The method comprises the following steps: Step S1: using a high-definition camera with large-angle shooting capability to cover multiple berth angles, and obtaining real-time images of the berths according to a preset shooting cycle to ensure that the images are complete and unobstructed; Step S2: Perform license plate location and recognition preprocessing on the vehicle area image detected in step S1. For license plates with large inclination angles, use a perspective transformation algorithm to correct the license plate inclination angle according to the mapping relationship between the parking lot ground calibration coordinates and the image pixel coordinates, so that the license plate is restored to a horizontal state; then perform character segmentation on the corrected license plate area, and segment a single character image according to the character spacing, width and connected domain characteristics, so as to prepare data for the subsequent character recognition stage; Step S3: License plate character recognition and traffic optimization: Input the segmented character image into the deep learning character recognition model. The model has been trained with a large number of license plate samples and can accurately recognize the character content and output the license plate number. Step S4: accurate inspection and abnormal judgment of berths, the flow-saving berth detection and inspection equipment starts the accurate inspection process, uses high-precision sensors installed at key positions of berths, conducts secondary inspections on suspected idle berths, accurately judges whether the berths are truly idle, and detects whether the parking posture of the vehicle is standardized; Step S5: Data fusion and parking space utilization analysis, integrating license plate recognition data with parking space inspection data to establish a real-time parking space usage database; Step S6: Equipment self-diagnosis and remote upgrade. The system regularly performs self-diagnosis on the high-angle license plate recognition equipment, flow-saving berth detection and inspection equipment, and checks the camera image acquisition quality, sensor working status, and communication module signal strength.

3. The method of a large inclination license plate recognition and berth management device according to claim 2, characterized in that: In step S4, if an abnormal situation is detected, such as misjudgment of the berth status or illegal parking of vehicles, an alarm message is immediately triggered, and the alarm signal is transmitted to the management center through the wireless communication module. The management center prominently displays the location of the abnormal berth on the monitoring large screen and pushes a processing task to the patrol terminal.

4. The method of a large - inclination license plate recognition and berth management device according to claim 3, characterized in that: In step S4, whether the parking posture of the vehicle is standard is calculated through the posture standardization detection formula: Assume that the parking space area is a rectangular area, and the position of the license plate area is [x1, y1, x2, y2], corresponding to the coordinates of its upper left corner and lower right corner. Then, whether the vehicle is pressing the line can be judged through the following inequality: Over = max(0, min(x2, x3) - max(x1, x4)) × max(0, min(y2, y3) - max(y1, y4)) [x3, y3, x4, y4] are the boundary coordinates of the berth area. If the overlapping part is less than a certain preset threshold, it is judged that the vehicle is not pressing the line.

5. The method for a large-inclination license plate recognition and berth management device according to claim 3, characterized in that: Assume that the normal state data value of the berth is S norm (i), if a value significantly different from the normal state is found in the sensor data, the abnormality degree A can be defined i (t), representing the degree of abnormality of berth i at time t, and the abnormality degree can be calculated by the following formula: A i (t) = ||D i (t) - S norm (i)|| / σ(D i ) Among them, σ(D i ) is the standard deviation of the sensor data of berth i, D i (t) represents the sensor data value. If A i (t) exceeds a certain threshold, it is considered that an abnormality has occurred at this berth.

6. The method of a large-inclination license plate recognition and berth management device according to claim 5, characterized in that: In step S5, based on this database, the parking space utilization rates in different time periods and different regions are analyzed, and a parking space utilization rate trend chart is drawn to provide a decision-making basis for the parking lot operation manager. For example, the charging strategy is reasonably adjusted according to peak and off-peak hours, and the layout of the parking space guiding signs is optimized. At the same time, potential demands are mined by data analysis, the parking space demand in future time periods is predicted, and preparations are made in advance to improve the overall operation efficiency of the parking lot.

7. A method for a large-inclination license plate recognition and berth management device according to claim 6, characterized in that: If the license plate positioning fails or the recognition accuracy rate is lower than the preset threshold, the image frame is automatically marked as abnormal, and a prompt message of "license plate recognition is abnormal, and the image needs to be manually verified" is generated and output. At the same time, the current processing flow is paused, waiting for manual intervention or re-acquisition of the image. If the license plate is recognized successfully and the accuracy rate meets the standard, the subsequent data processing flow is continued, and the valid data is stored in the database.

8. A method for a large-inclination license plate recognition and berth management device according to claim 7, characterized in that: In step S2, the perspective transformation algorithm is: The position of the license plate in the three-dimensional space is (X, Y, Z), and the image coordinates (x, y) are obtained through perspective transformation. The expression of the perspective transformation matrix H is: Among them, H is the perspective transformation matrix; The character segmentation is specifically based on the deep learning Mask R-CNN or U-Net segmentation network to segment characters, and obtain the region R of each character c , and the goal of the model is to optimize the segmentation accuracy by minimizing the intersection over union (IoU) loss: Among them, is the intersection over union between the real character region and the predicted region, is the character region predicted by the model; The character recognition uses a convolutional neural network or a recurrent neural network to recognize the segmented characters. Let the segmented character image be I(x, y). The character recognition model is trained through deep learning and outputs the character recognition result C i , and the loss function for training is as follows: Among them, y is the true character label, and p i is the character probability predicted by the model. N is the number of characters. The smaller the loss value is, the closer the prediction result of the model is to the true label.

Images