A patrol vehicle parking space positioning system and method based on OCR

Abstract

The present invention discloses a patrol vehicle parking space positioning system and method based on OCR, which relates to the field of intelligent transportation technology. The system includes: an image acquisition module, which is designed to capture parking space images through a camera; a vehicle positioning module, which is designed to obtain positioning information of a video patrol vehicle through a GPS positioning device; a data processing module, which is designed to pre-process parking space images and analyze image data according to image algorithms; and a mobile communication module, which is designed to transmit license plate and parking space information to an operation management platform through mobile communication technology. According to the technical solution of the present application, the function of quickly identifying vehicle license plate information and matching parking spaces in parking fee scenarios can be realized, which has a high application value.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to an OCR-based patrol vehicle parking space positioning system and method. Background Art

[0002] In recent years, the number of motor vehicles in my country has continued to grow, and cars have become an indispensable means of transportation for people's daily travel. The problem of parking difficulties has become increasingly prominent. Especially in busy commercial areas, residential areas and near public transportation hubs, finding a suitable parking space has become a headache for many drivers. In order to improve the utilization rate of on-street parking spaces, many intelligent road inspection vehicles have appeared on the market, which reduce manual participation and improve the efficiency of road inspections. Video inspection vehicles are used to collect parking conditions of on-street parking and take photos of vehicles in parking spaces for evidence collection. Among them, matching vehicles to corresponding parking spaces is one of the core functions of video inspection vehicles and is an important part that affects the accuracy of their inspections. At present, most of the existing parking space matching solutions are to calibrate the static GPS latitude and longitude data of each parking space in advance based on the real-time GPS data received by the GPS device with the help of high-frequency and high-precision GPS equipment, so as to complete the parking space matching work. However, this solution still has some significant problems: before the video inspection vehicle is put into use, the latitude and longitude data of the starting point and end point of each berth need to be collected. The workload of the collection process is extremely heavy, and the amount of data required to be collected is extremely large; for environments with weak satellite positioning signals, such as high-rise buildings, tree obstructions, signal shielding, etc., the inaccuracy of GPS positioning will cause the latitude and longitude data of the pre-berth calibration to deviate from the latitude and longitude data of the actual berth, thus affecting the normal operation of the video inspection vehicle. Therefore, the development of a simple and efficient inspection vehicle parking location positioning method is imminent.

[0003] Most of the existing methods for patrol car parking space positioning currently have defects in some aspects. For example, the solution of NB geomagnetism combined with video patrol car, although the device layout is convenient and has the advantages of high precision and low false alarm rate, the dual-mode geomagnetism has insufficient performance stability under special weather conditions, and the three-mode geomagnetism has the problem of high cost; the SLAM solution, although it can estimate the posture of the moving vehicle in real time and build a map of the surrounding environment, but the visual SLAM cannot work in the dark or textureless area, and the application scenarios of laser SLAM are relatively limited. Therefore, based on this problem, this application proposes a patrol car parking space positioning system and method based on OCR. Summary of the invention

[0004] Technical Purpose

[0005] In order to solve the above problems, the purpose of the present invention is to provide an OCR-based patrol vehicle parking positioning system and method, which not only solves the time-consuming and labor-intensive problem of manually collecting berth longitude and latitude data in the traditional berth positioning process, but also avoids the occurrence of inaccurate longitude and latitude data due to human factors, realizes the simplification and automation of berth positioning, significantly improves the inspection accuracy of video patrol vehicles and their adaptability to different scenarios, and avoids the application obstacles of video patrol vehicles caused by objective factors such as satellite positioning.

[0006] Technical Solution

[0007] In order to achieve the above objectives, the present invention provides an OCR-based inspection vehicle parking space positioning system and method, which recognizes the parking space numbers sprayed on the ground through image segmentation and OCR text recognition algorithm, thereby achieving accurate matching of on-street parking spaces.

[0008] In a first aspect, the present invention provides an OCR-based inspection vehicle parking space positioning system, comprising:

[0009] An image acquisition module, which acquires a parking space image through a camera, wherein the parking space image includes a vehicle license plate number and a parking space number;

[0010] Vehicle positioning module, which obtains the positioning information of the video inspection vehicle through GPS positioning equipment;

[0011] A data processing module pre-processes the parking space image and analyzes the image data according to an image algorithm to obtain the license plate information and the parking space information, and determines the parking space corresponding to the captured image by the time of the picture frame, the time corresponding to the license plate shooting, and the position of the parking space number in the image; wherein the image algorithm includes an OCR text recognition algorithm;

[0012] The mobile communication module transmits license plate and parking space information to the operation management platform through mobile communication technology.

[0013] Furthermore, the vehicle's license plate number is captured and identified in real time by an intelligent license plate recognition camera installed on the vehicle inspection vehicle; the berth number is captured and identified in real time by a special camera installed on the vehicle inspection vehicle for collecting a clear berth number video stream; the intelligent license plate recognition camera and the special camera are installed at the front end of the video inspection vehicle according to the relative positions of the vehicle's license plate and the berth number.

[0014] Furthermore, after receiving the video stream collected by the berth number collection camera, the data processing module obtains the berth number data through the built-in OCR text recognition algorithm of the data processing module, which pre-processes the picture frames in the video stream, segments the berth number characters in the picture frames, and recognizes the content of the berth number characters, thereby obtaining the berth number data. It not only solves the problem of time-consuming and labor-intensive manual collection of berth longitude and latitude data in the traditional berth positioning process, but also avoids the occurrence of inaccurate longitude and latitude data due to human factors, realizes the simplification and automation of berth positioning, significantly improves the inspection accuracy of the video inspection vehicle and its adaptability to different scenarios, and avoids the application obstacles of the video inspection vehicle caused by objective factors such as satellite positioning.

[0015] Furthermore, the data processing module preprocesses the picture frames in the video stream, specifically including: marking text areas, determining the location of the text in each document by marking the text areas of the collected sample data, thereby building a model to learn the spatial structure of the text and improve the recognition accuracy of the text; extracting text content, extracting the text content in the marked text area and establishing label data corresponding to the text content for supervised learning; data cleaning and preprocessing, cleaning and preprocessing the extracted text content, including removing noise, adjusting size and direction, etc., to ensure the quality of the text data.

[0016] Furthermore, the OCR text recognition algorithm supports multiple models, including traditional template matching algorithms, support vector machines, Bayesian classification algorithms, and neural network models based on deep learning, etc., and it can select appropriate models for training according to specific application scenarios and needs; after the model training is completed, the data processing module improves the accuracy of the OCR text recognition algorithm through data set optimization, feature selection, and model structure optimization, wherein the data set optimization specifically uses a more accurate and comprehensive data set for training, and optimizes the data set by increasing the sample size, deleting duplicate samples, increasing randomness, etc.; feature selection specifically selects the most important features for the recognition task and reduces unnecessary features, thereby speeding up the response speed of the OCR system, and the selection of relevant features is based on feature evaluation methods, such as information gain, correlation coefficient, etc.

[0017] Furthermore, the data processing module improves the recognition accuracy and speed of the OCR text recognition algorithm based on a deep learning solution, which uses a convolutional neural network as a basic model and continuously improves the recognition ability of the OCR text recognition algorithm by increasing the amount of training data; the deep learning solution includes data preprocessing, building a convolutional neural network model, calculating loss functions, back propagation and optimization, regularization, etc. It can help the OCR system better process text in complex environments, improve the recognition accuracy of the OCR system for text under non-ideal conditions, significantly speed up the model's operation speed, and reduce the need for manual feature engineering by automatically learning features in images, thereby simplifying the development process.

[0018] Furthermore, the spraying of the berth number is based on a customized OCR system-specific number spraying standard, which involves the font, size and color of the berth number, so as to improve the applicability of the berth number OCR algorithm. It helps the OCR system to more accurately identify the berth number, reduces the time required for the OCR algorithm to process the image, enables the OCR system to more easily adapt to different environmental conditions, and improves the recognition ability of the OCR system in a variety of scenarios.

[0019] Furthermore, the data processing module determines whether the berth information of the current berth is correct by comprehensively comparing the berth information and the positioning information of the video inspection vehicle of the previous berth with the berth information and the positioning information of the video inspection vehicle of the current berth; for example, if the berth number of the previous berth is 00050, the physical distance between the positioning of the video inspection vehicle of the previous berth and the positioning of the video inspection vehicle of the current berth is only 6 meters, but the berth number of the current berth is calculated to be 00031, then it is determined that the berth number of the current berth may not be correctly identified, and the berth number of the current berth can be corrected according to the calculation. The berth information of the next berth and the positioning information of the video inspection vehicle are comprehensively verified. If the video inspection vehicle recognizes that the number of the next berth is 00052 after moving only about 6 meters, it can be basically concluded that the actual number of this berth should be 00051. The reason for failing to correctly identify the berth number of this berth may be that the berth number is blocked, contaminated or damaged. Similar abnormal situations can be submitted to the operation management platform for subsequent processing, and the image algorithm can be assisted from multiple angles to judge the license plate information and berth information, improve the recognition and positioning accuracy, and improve the operation management efficiency.

[0020] Furthermore, the mobile communication module establishes a parking space geographic information database based on the geographic information solution, which displays the location distribution map of parking spaces in the form of a map, including: matching the data recognized by the OCR system with the geographic information in the geographic information solution database, which should contain information such as the geographic coordinates, belonging area, and type of the parking space; combining the real-time collected data with the geographic information solution, and performing spatial analysis and processing, using the geographic information solution to monitor and manage the vehicle position and parking space status in real time; displaying the parking space status, vehicle position, and related statistical information through a map view. It can quickly respond to parking problems by monitoring the parking space status on the road and the flow of nearby vehicles in real time, and can more effectively allocate and manage parking space resources, provide drivers with fast and accurate parking space information, and reduce the time drivers spend looking for parking spaces.

[0021] In a second aspect, the present invention further provides a method for locating parking spaces for patrol vehicles based on OCR, the method being based on the system described in the first aspect, comprising:

[0022] The berth image is obtained through the camera installed on the video inspection vehicle;

[0023] Get the positioning information of the video inspection vehicle through GPS positioning equipment;

[0024] Preprocessing the berth image and analyzing the image data according to the image algorithm;

[0025] Output license plate and parking space information through mobile communication technology.

[0026] In a third aspect, the present invention further provides a computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the aforementioned OCR-based inspection vehicle parking space positioning method is implemented.

[0027] The present invention identifies the berth number sprayed on the ground through image segmentation and OCR text recognition algorithm. The spraying of the berth number is based on the number spraying standard dedicated to the customized OCR system; and a parking space geographic information database is established based on the geographic information solution, and the location distribution map of the parking space is displayed in the form of a map, and the recognition accuracy and speed of the OCR text recognition algorithm are improved based on the deep learning solution. The system and method not only solve the problem of time-consuming and labor-intensive manual collection of berth longitude and latitude data in the traditional berth positioning process, but also avoids the occurrence of inaccurate longitude and latitude data due to human factors, realizes the simplification and automation of berth positioning, significantly improves the inspection accuracy of the video inspection vehicle and its adaptability to different scenes, and avoids the application barrier problem of the video inspection vehicle caused by objective factors such as satellite positioning.

[0028] Beneficial Effects

[0029] By implementing the OCR-based inspection vehicle parking space positioning system and method provided by the present invention, the following technical effects are achieved:

[0030] (1) The present invention recognizes the berth number sprayed on the ground through image segmentation and OCR text recognition algorithm; it not only solves the time-consuming and labor-intensive problem of manually collecting berth longitude and latitude data in the traditional berth positioning process, but also avoids the occurrence of inaccurate longitude and latitude data due to human factors, realizes the simplification and automation of berth positioning, significantly improves the inspection accuracy of the video inspection vehicle and its adaptability to different scenarios, and avoids the application obstacles of the video inspection vehicle caused by objective factors such as satellite positioning.

[0031] (2) The spraying of the berth number of the present invention is based on a customized number spraying standard dedicated to the OCR system; it helps the OCR system to more accurately identify the berth number, reduces the time required for the OCR algorithm to process the image, enables the OCR system to more easily adapt to different environmental conditions, and improves the recognition ability of the OCR system in a variety of scenarios.

[0032] (3) The present invention establishes a parking space geographic information database based on a geographic information solution and displays the location distribution of parking spaces in the form of a map; it quickly responds to parking problems by real-time monitoring the status of on-street parking spaces and the flow of nearby vehicles, and can more effectively allocate and manage parking space resources, providing drivers with fast and accurate parking space information and reducing the time drivers spend looking for parking spaces.

[0033] (4) The present invention improves the recognition accuracy and speed of the OCR text recognition algorithm based on a deep learning solution, and adopts a convolutional neural network as the basic model; it can help the OCR system better process text in complex environments, improve the recognition accuracy of the OCR system for text under non-ideal conditions, significantly speed up the calculation speed of the model, and reduce the need for manual feature engineering by automatically learning the features in the image, thereby simplifying the development process. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to make the above-mentioned OCR-based inspection vehicle parking space positioning system and method of the present invention more obvious and easy to understand, the following is a brief introduction to the drawings required for use in the specific implementation of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without paying creative labor.

[0035] Figure 1 It shows the principle diagram of the parking space positioning system for patrol vehicles;

[0036] Figure 2 A schematic diagram showing the location of the camera for collecting berth numbers;

[0037] Figure 3 Schematic diagram showing the standard for spraying berth numbers;

[0038] Figure 4 A schematic diagram showing the effect of the OCR text recognition algorithm;

[0039] Figure 5 A schematic diagram showing the effect of the geographic information solution. DETAILED DESCRIPTION

[0040] Embodiment 1:

[0041] Provides an OCR-based inspection vehicle parking space positioning system, such as Figure 1 As shown, the inspection vehicle is equipped with a GPS receiver to receive signals from satellites, which include timestamps and satellite location information; the base station is a fixed GPS receiver whose location is known. The base station receives the same satellite signal and sends the received data to the inspection vehicle through radio broadcasts; the radio, as a communication medium, transmits the correction information received by the base station to the inspection vehicle in real time; after receiving the correction information broadcast by the base station, the PTK positioning module on the inspection vehicle uses this information to correct the original GPS signal to eliminate or reduce the error of the satellite signal, such as ionospheric delay, tropospheric delay, etc.; after PTK processing, the inspection vehicle can obtain higher positioning accuracy than using GPS alone. The inspection vehicle integrates the precise location information obtained by the PTK positioning module with other sensor data to achieve more accurate parking space positioning; finally, the inspection vehicle transmits the positioning information as well as the license plate and parking space information to the operation management platform through the mobile communication module for final processing and analysis. The system includes: an image acquisition module, which is designed to obtain a parking space image through a camera, wherein the parking space image includes the vehicle's license plate number and parking space number; a vehicle positioning module, which is designed to obtain the positioning information of the video inspection vehicle through a GPS positioning device; a data processing module, which is designed to pre-process the parking space image and analyze the image data according to an image algorithm to obtain the license plate information and parking space information, and judge the parking space corresponding to the captured image through the time of the picture frame, the time corresponding to the license plate shooting, and the position of the parking space number in the image, wherein the image algorithm includes an OCR text recognition algorithm; a mobile communication module, which is designed to transmit the license plate and parking space information to the operation management platform through mobile communication technology.

[0042] The OCR-based inspection vehicle parking space positioning system provided by the present invention specifically includes:

[0043] The vehicle's license plate number is captured and identified in real time by an intelligent license plate recognition camera installed on the vehicle inspection vehicle; the berth number is captured and identified in real time by a special camera installed on the vehicle inspection vehicle for collecting a clear berth number video stream; the intelligent license plate recognition camera and the special camera are installed at the front end of the video inspection vehicle according to the relative positions of the vehicle's license plate and the berth number.

[0044] After receiving the video stream collected by the berth number collection camera, the data processing module obtains the berth number data through the built-in OCR text recognition algorithm of the data processing module. It pre-processes the picture frames in the video stream, segments the berth number characters in the picture frames, and recognizes the content of the berth number characters, thereby obtaining the berth number data. The position of the berth number collection camera is as follows: Figure 2 shown.

[0045] The data processing module preprocesses the picture frames in the video stream, specifically including: marking text areas, determining the location of the text in each document by marking the text areas of the collected sample data, thereby building a model to learn the spatial structure of the text and improve the recognition accuracy of the text; extracting text content, extracting the text content in the marked text area and establishing label data corresponding to the text content for supervised learning; data cleaning and preprocessing, cleaning and preprocessing the extracted text content, including removing noise, adjusting size and direction, etc., to ensure the quality of the text data.

[0046] The OCR text recognition algorithm supports multiple models, including traditional template matching algorithms, support vector machines, Bayesian classification algorithms, and neural network models based on deep learning, etc. It can select appropriate models for training according to specific application scenarios and needs; after the model training is completed, the data processing module improves the accuracy of the OCR text recognition algorithm through data set optimization, feature selection, and model structure optimization, wherein the data set optimization specifically uses a more accurate and comprehensive data set for training, and optimizes the data set by increasing the sample size, deleting duplicate samples, increasing randomness, etc.; feature selection specifically selects the most important features for the recognition task and reduces unnecessary features, thereby speeding up the response speed of the OCR system, and the selection of relevant features is based on feature evaluation methods, such as information gain, correlation coefficient, etc.

[0047] The berth number is sprayed based on a custom OCR system-specific number spraying standard, which involves the font, size and color of the berth number to improve the applicability of the berth number OCR algorithm. The berth number spraying standard is as follows: Figure 3 As shown, the effect of the OCR text recognition algorithm is as follows Figure 4 shown.

[0048] The data processing module determines whether the berth information of the current berth is correct by comprehensively comparing the berth information and the positioning information of the video inspection vehicle of the previous berth with the berth information and the positioning information of the video inspection vehicle of the current berth; for example, if the berth number of the previous berth is 00050, and the physical distance between the positioning of the video inspection vehicle of the previous berth and the positioning of the video inspection vehicle of the current berth is only 6 meters, but the berth number of the current berth is calculated to be 00031, then it is determined that the berth number of the current berth may not be correctly identified, and the berth number of the next berth may be used to identify the berth number of the current berth. If the video inspection vehicle recognizes the next berth as 00052 after moving only about 6 meters, it can be basically concluded that the actual number of this berth should be 00051. The reason for failing to correctly identify the berth number of this berth may be that the berth number is blocked, contaminated or damaged. Similar abnormal situations can be submitted to the operation management platform for subsequent processing, which can assist the image algorithm in judging the license plate information and berth information from multiple angles, improve the recognition and positioning accuracy, and improve the operation management efficiency.

[0049] The mobile communication module establishes a parking space geographic information database based on the geographic information solution, which displays the location distribution map of parking spaces in the form of a map, including: matching the data recognized by the OCR system with the geographic information in the geographic information solution database, which should contain the geographic coordinates, belonging area, type and other information of the parking space; combining the real-time collected data with the geographic information solution, and performing spatial analysis and processing, using the geographic information solution to monitor and manage the vehicle location and parking space status in real time; and displaying the parking space status, vehicle location and related statistical information through a map view. The effect of the geographic information solution is as follows: Figure 5 shown.

[0050] Embodiment 2:

[0051] On the basis of the above-mentioned embodiment, a deep learning scheme is added to improve the recognition accuracy and speed of the OCR text recognition algorithm, which adopts a convolutional neural network as a basic model and continuously improves the recognition ability of the OCR text recognition algorithm by increasing the number of training data;

[0052] Among them, it is necessary to preprocess the data in advance and normalize the image data to a uniform range, such as or , ;

[0053] in, Represents the original image pixel value; Indicates the minimum value of the image pixel value, usually 0; Indicates the maximum value of the image pixel value, usually 255; Represents the normalized pixel value, ranging from 0 to 1.

[0054] Building a convolutional neural network involves:

[0055] Multiple convolution kernels are applied in the convolution layer to extract image features:

[0056]

[0057] in, Represents the convolution kernel Output: Represents the weight of the convolution kernel; Represents the pixel value of the input image; Indicates bias; Represents the total number of convolution kernels;

[0058] use As a non-linear activation function:

[0059]

[0060] Reduce the dimensionality of the feature map by using max pooling:

[0061]

[0062] in, Represents an element on the feature map output by the pooling layer; represents the elements on the input feature map, and Defines the position within the pooling window;

[0063] Flatten the feature map and connect it to the output layer:

[0064]

[0065] in, Indicates The output of a neuron; represents weight; Represents input features; represents the bias term of the fully connected layer; Indicates the length of the input feature vector.

[0066] Loss function for multi-classification problems:

[0067]

[0068] in, Represents the value of the loss function, which measures the difference between the model prediction and the true label; The encoding representing the true label; represents the probability distribution predicted by the model; Indicates the total number of categories.

[0069] Calculate the gradient of the loss function with respect to the model parameters:

[0070]

[0071] in, Represents the loss function with respect to weights The gradient of Represents the loss function with respect to the fully connected layer output The gradient of Represents the activation function with respect to its input The gradient of Represents the fully connected layer input About weight The gradient of

[0072] And update the model parameters using gradient descent:

[0073]

[0074] in, represents the learning rate; Represents the weight in the next iteration; Represents the weight in the current iteration.

[0075] Prevent overfitting through L2 regularization:

[0076]

[0077] in, Represents the regularization coefficient, which controls the strength of the regularization term.

[0078] For example, there is a dataset containing 1000 pictures of parking lot numbers, each picture is of size Pixels, the image is grayscale, the image has been preprocessed and normalized to Range, berth number has 10 possible digits, from 0 to 9;

[0079] Suppose there is a berth number map , whose pixel values ​​range from 0 to 255, normalize them to scope:

[0080]

[0081] Build a convolutional neural network model, including:

[0082] Convolution layer: 32 convolution kernels, size , using ReLU activation function;

[0083] Pooling layer: maximum pooling, window size is ;

[0084] Fully connected layer: 128 neurons, also using ReLU activation function;

[0085] Output layer: 10 neurons, corresponding to 10 digital categories;

[0086] Using the cross entropy loss function, assuming the true label is , the model predicts :

[0087]

[0088] Assuming a learning rate of 0.01, after a forward propagation and loss calculation, the gradient is calculated and the weights are updated:

[0089]

[0090]

[0091] L2 regularization is used with a regularization coefficient of 0.001:

[0092]

[0093]

[0094] The effect of the deep learning solution is shown in Table 1:

[0095] Table 1. Summary of deep learning solution effects

[0096] Training cycle Training loss Training accuracy Validation loss Verification accuracy 1 2.30 0.20 2.35 0.18 2 1.50 0.45 1.60 0.40 3 1.10 0.60 1.25 0.55 4 0.90 0.70 1.10 0.60 5 0.75 0.75 0.95 0.65 6 0.65 0.80 0.85 0.70 7 0.55 0.85 0.75 0.75

[0097] As shown in the table, as training progresses, the training loss gradually decreases and the training accuracy gradually increases, which indicates that the performance of the model on the training data is improving; at the same time, the validation loss and validation accuracy also show that the performance of the model on unseen data is improving, but the improvement is not as large as that on the training set.

[0098] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable non-transient storage media containing computer-usable program code.

[0099] The present invention can provide computer program instructions to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce means for implementing the system.

[0100] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions of the system.

[0101] These computer program instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions of the system.

Claims

1. An OCR-based inspection vehicle parking space positioning system, characterized by: include: The image acquisition module is designed to acquire a berth image through a camera, wherein the berth image Include the vehicle's license plate number and parking space number; The vehicle positioning module is designed to obtain the positioning information of the video inspection vehicle through the GPS positioning device; The data processing module is designed to pre-process the parking space image and analyze the image data according to the image algorithm to obtain the license plate information and the parking space information, and determine the parking space corresponding to the captured image according to the time of the picture frame, the time corresponding to the license plate shooting and the position of the parking space number in the image; wherein the image algorithm includes an OCR text recognition algorithm; The mobile communication module is designed to transmit license plate and parking space information to the operation management platform through mobile communication technology; The data processing module improves the recognition accuracy and speed of the OCR text recognition algorithm based on a deep learning solution. It uses a convolutional neural network as a basic model and continuously improves the recognition ability of the OCR text recognition algorithm by increasing the amount of training data. It is necessary to pre-process the data in advance and normalize the image data to a unified range. ,in, Represents the original image pixel value; Indicates the minimum value of the image pixel value, usually 0; Indicates the maximum value of the image pixel value, usually 255; Represents the normalized pixel value, ranging from 0 to 1; Building a convolutional neural network involves: Multiple convolution kernels are applied in the convolution layer to extract image features: ,in, Represents the convolution kernel Output: Represents the weight of the convolution kernel; Represents the pixel value of the input image; Indicates bias; Represents the total number of convolution kernels; use As a non-linear activation function: Reduce the dimensionality of the feature map by using max pooling: ,in, Represents an element on the feature map output by the pooling layer; represents the elements on the input feature map, and Defines the position within the pooling window; Flatten the feature map and concatenate it to the output layer: ,in, Indicates The output of a neuron; represents weight; Represents input features; represents the bias term of the fully connected layer; Represents the length of the input feature vector; Loss function for multi-classification problems: ,in, Represents the value of the loss function, which measures the difference between the model prediction and the true label; The encoding representing the true label; represents the probability distribution predicted by the model; Indicates the total number of categories; Calculate the gradient of the loss function with respect to the model parameters: ,in, Represents the loss function with respect to weights The gradient of Represents the loss function with respect to the fully connected layer output The gradient of Represents the activation function with respect to its input The gradient of Represents the fully connected layer input About weight The gradient of And update the model parameters using gradient descent: ,in, represents the learning rate; Represents the weight in the next iteration; Represents the weight in the current iteration; Prevent overfitting through L2 regularization: ,in, Represents the regularization coefficient, which controls the strength of the regularization term.

2. The OCR-based inspection vehicle parking space positioning system according to claim 1 is characterized in that: The vehicle's license plate number is captured and identified in real time by an intelligent license plate recognition camera installed on the vehicle inspection vehicle; the berth number is captured and identified in real time by a special camera installed on the vehicle inspection vehicle for collecting a clear berth number video stream; the intelligent license plate recognition camera and the special camera are installed at the front end of the video inspection vehicle according to the relative positions of the vehicle's license plate and the berth number.

3. The OCR-based inspection vehicle parking space positioning system according to claim 1 or 2, characterized in that: After receiving the video stream captured by the berth number acquisition camera, the data processing module obtains the berth number data through the OCR text recognition algorithm built into the data processing module. It obtains the berth number data by pre-processing the picture frames in the video stream, segmenting the berth number characters in the picture frames, and identifying the content of the berth number characters.

4. The OCR-based inspection vehicle parking space positioning system according to claim 3 is characterized in that: The data processing module preprocesses the picture frames in the video stream, specifically including: marking text areas, determining the location of text in each document by marking text areas in the collected sample data; extracting text content, extracting text content in the marked text area and establishing label data corresponding to the text content; data cleaning and preprocessing, cleaning and preprocessing the extracted text content, including removing noise, adjusting size and direction.

5. The OCR-based inspection vehicle parking space positioning system according to claim 3 is characterized in that: The OCR text recognition algorithm supports multiple models, including traditional template matching algorithms, support vector machines, Bayesian classification algorithms, and neural network models based on deep learning. It can select appropriate models for training according to specific application scenarios and needs. After the model training is completed, the data processing module improves the accuracy of the OCR text recognition algorithm through data set optimization, feature selection, and model structure optimization.

6. The OCR-based inspection vehicle parking space positioning system according to claim 1 is characterized in that: The spraying of the berth number is based on a number spraying standard dedicated to a customized OCR system, and the number spraying standard involves the font, font size and color of the berth number.

7. The OCR-based inspection vehicle parking space positioning system according to claim 1 is characterized in that: The mobile communication module establishes a parking space geographic information database based on a geographic information solution, which displays the location distribution diagram of parking spaces in the form of a map.

8. A method for locating parking spaces for patrol vehicles based on OCR, characterized in that: The method is implemented based on the system according to any one of claims 1 to 7: The method comprises: The berth image is obtained through the camera installed on the video inspection vehicle; Get the positioning information of the video inspection vehicle through GPS positioning equipment; Preprocessing the berth image and analyzing the image data according to the image algorithm; Output license plate and parking space information through mobile communication technology.

9. A computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, characterized in that: When the computer program is executed, the OCR-based inspection vehicle parking space positioning method according to claim 8 is implemented.

Images