CN116129416A - AI algorithm-based vehicle management system with double systems and double modes - Google Patents

Abstract

The invention discloses a dual-system dual-mode vehicle management system based on an AI algorithm, which comprises the construction of a data set, the training and compression construction of a model and the design of a vehicle management system platform, adopts a plurality of algorithm complementation modes, and provides a dual-mode dual-system vehicle information management system which is suitable for most application scenes in the whole country, thereby meeting most of the demands, solving the problems that the entrance and exit efficiency of most of the large-scale parking lots in the whole country is low, users are easy to lose directions and other pain points in the parking lots, simultaneously solving the problems that the vehicle recognition accuracy is low and the vehicle recognition accuracy is low when the air visibility is low in a haze weather frequent region, solving the problem that the vehicle recognition accuracy of coal enterprises in recognizing transportation coal trucks shielded by coal slime is low, meeting the demands of industrial production and the demands of common people on individuation, reducing unnecessary time waste, and becoming an effective tool for the convenience of people.

Description

AI algorithm-based vehicle management system with double systems and double modes

Technical Field

The invention belongs to the technical field related to vehicle management systems, and particularly relates to a vehicle management system with double systems and double modes based on an AI algorithm.

Background

As deep learning advances in wave, deep learning has been increasingly applied to infrastructure, slowly penetrating into various industries and fields; the traditional character recognition needs to manually extract features, and the steps are complicated; the maturing of the deep convolutional neural network technology and the improvement of the computing power of edge equipment lead the license plate recognition technology to be applied;

at present, there is no more comprehensive vehicle management system in the domestic market, and most license plate recognition technologies of the parking management system cannot be applied to such as: scenes such as large haze weather, coal transportation sites and the like; in a haze weather frequent area, when the air visibility is low, the vehicle identification accuracy is low and the speed is low; when a coal enterprise is involved, the accuracy of identifying license plates of transportation coal trucks shielded by coal slime is low; in addition, most large parking lots in the whole country have low entrance and exit efficiency, and users are easy to lose directions in the parking lots; the industrial production requirement and the personalized requirement of the common people cannot be met, unnecessary time waste is caused, and the common people cannot be facilitated.

Disclosure of Invention

The invention aims to provide a dual-system dual-mode vehicle management system based on an AI algorithm, which is used for solving the problems that the vehicle recognition accuracy is low, the speed is low, the entrance and exit efficiency of a large parking lot is low, users are easy to lose directions in the parking lot, unnecessary time waste is caused, and the convenience is brought to people.

In order to achieve the above purpose, the present invention provides the following technical solutions: the vehicle management system with double systems and double modes based on the AI algorithm comprises the construction of a data set, the training and compression construction of a model and the design of a vehicle management system platform;

construction of a data set: the method has the advantages that various data sets with clear targets serving as license plate recognition are supplemented by using ImageNet, MSCOCO, PASCAVOC, CCPD and other public data sets and the principle of team self-shooting, and the Internet and the inside of the team have better image preprocessing processes for the data sets used by the method;

training and compression of the model: the identification of the vehicle identity adopts a vehicle identification technology, and the automatic identification of the license plate is a core function of the whole vehicle management system; different vehicle recognition algorithms are adopted for the vehicle recognition modes under different environments, wherein the vehicle recognition modes under different environments are as follows: a conventional environment rapid identification mode, a haze weather mode and a coal enterprise management system;

design of a vehicle management system platform: the method is divided into main functions and architecture design of a conventional small and medium-sized parking lot management system, main functions and architecture design of a coal enterprise management system and database design.

Preferably, in the construction of the data set, after enough pictures are collected, a marking tool LabelImg can be used for marking license plate pictures; wherein, chinese characters cannot be marked directly, none is used for replacing the Chinese characters, and Unicode codes are used for distinguishing the Chinese characters manually in the later period; the three modes comprise three different data sets, and taking a haze weather mode as an example, the haze weather mode data sets are constructed into a VOC data format; firstly, creating a file named Voc, and then creating Label_list. Txt, tranval. Txt, test. Txt and VOCdevkit file in the Voc file; wherein the Label_list file stores class labels of training data; trainoval. Txt and test. Txt store pictures of training data and test data under the relative Voc file path and the corresponding paths of the labeling files; newly building a VOC2022 file in the VOCdevkit folder, and newly building a Annotations, imageSets, JPEGImage folder in the VOC 2022; the announcements folder stores license plates and license plate character labeling files; in Main files in ImageSets, the picture names of the test data and the training data are stored; the JPEGImages folder stores collected license plate pictures; the Labels file stores a marking file in a Yolo format and a VOC data set file organization chart.

Preferably, the conventional context quick recognition mode: aiming at the key points and difficulties that the license plate detection and recognition algorithm needs to be realized in the conventional environment, the license plate detection and recognition algorithm can be divided into three modules: license plate detection, image processing and OCR license plate number recognition;

(1) And (3) license plate detection: YOLOv4 target detection model based on improved SOTA method:

the yolov4 target detection algorithm consists of five basic components: CBM (conv+bn+mix activation function), CBL (conv+bn+leak_relu activation function), resunit (residual structure), CSPX (convolutional layer and Resunit module), SPP (pooling layer); the method has the main advantages that yolov4 is an efficient and powerful target detection model, and can train an ultrafast and accurate target detector by using relatively moderate computing power equipment; during the training process of the detector, the influence of some most advanced research results on the target detector is verified; the SOTA method is improved, so that the SOTA method is more effective and more suitable for single GPU training;

(2) Image processing: image noise reduction and binarization processing based on Opencv:

in order to further improve the recognition accuracy, the change of the light condition is that a certain difference exists in the detected and segmented image under normal conditions, so that the image is simply subjected to noise reduction and binarization processing by using an Opencv library, and the image is more suitable for OCR (optical character recognition) to be further recognized;

(3) OCR license plate number recognition: hundred-degree flying-oar OCR-based open-source text recognition:

the OCR is text recognition based on image detection, has the advantages of simple model training, small volume, high speed and high accuracy, and can quickly obtain license plate numbers through recognition processed license plate images in a license plate recognition system, thereby realizing quick vehicle entering and exiting of parking lots.

Preferably, haze weather pattern: aiming at key points and difficulties which need to be realized by a license plate detection and recognition algorithm in haze weather, the license plate detection and recognition algorithm can be divided into three modules: defogging images, positioning license plates and recognizing license plate characters;

(1) Image defogging: based on a prior defogging algorithm for fast guiding and filtering dark primary colors;

analyzing and comparing the decontamination effects of an image restoration defogging algorithm, a dark primary prior defogging algorithm and a fast guiding and filtering dark primary prior defogging algorithm based on an atmospheric scattering model, and selecting the defogging algorithm with the optimal effect based on the fast guiding and filtering dark primary prior defogging algorithm as a haze weather mode of the project;

(2) License plate positioning: license plate positioning algorithm based on edge detection;

edge detection is a common method in digital image processing, and the algorithm is to identify points with obvious brightness change in an image so as to realize edge contour distinction; in the license plate recognition system, the color difference between the color of the vehicle body and the color of the character is larger, so that the pixel lighting brightness jump at the edge of the license plate is large, and the outline of the edge is clear and is convenient to recognize; firstly, carrying out preprocessing such as denoising, gray stretching and the like on a license plate image to obtain an image with clear edge contour, counting the gray value of each pixel point in the image, and finding out a point with large gray value change, which is called a gray image jump point; the license plate position can be positioned according to the number of the jump points, the algorithm can quickly and effectively identify the license plate of the vehicle in the actual application process, the realization difficulty is low, and the practical cost performance is extremely high;

(3) License plate character recognition: YOLOv3 identification network based on MobileNetV3 network structure;

training a haze weather license plate data set in a common target detection algorithm, and finding that a MobileNet V3 network structure-based Yolov3 recognition network algorithm compressed by a sensitivity-based method is selected to recognize a fuzzy license plate, so that higher accuracy can be ensured, a certain instantaneity is realized, and the requirements of real tasks can be met; and the realization difficulty of the recognition network algorithm is low, so the algorithm is adopted as a core algorithm of the license plate character recognition process of the accurate mode of the haze weather of the project.

Preferably, the coal enterprise management system is an extension of the haze weather mode, and a license plate character recognition algorithm trained by the same data set as the haze weather mode but different data sets is used (the algorithm has great advantage in the aspect of processing tiny shielding objects, and the influence of haze weather and coal slime on license plates has various commonalities, so that the effect is inferred to be feasible), and the difference is that the design and implementation part of the vehicle management system platform are explained in detail.

Preferably, the main functions and the architecture design of the conventional small and medium-sized parking lot management system are as follows:

(1) And (3) main functional analysis: the service scene of the conventional medium and small parking lot vehicle information management system and the use requirement of a user on products are combined, and the functional requirement of the system is analyzed and summarized by the service scene, which mainly comprises the following points:

(1) providing basic data management functions for the administrator user by the system, including user management, role allocation and department management; (2) providing a set of perfect automatic charging flow and charging information management system which can be controlled by a manager; (3) the system should have the system information monitoring functions such as log of user, operation log and the like; (4) the administrator user can manage the cameras according to actual needs, such as modifying camera information, deleting cameras and the like; (5) the user can preview the picture shot by the camera, set the detection area, check the detection state and whether to start detection; (6) the user can inquire specific information of the vehicle detected from the monitoring video according to the time, the camera name and other conditions, such as license plate color, number and snap-shot pictures; (7) the system can acquire real-time video stream through the video stream address filled by the user, then carries out vehicle identification, calls a license plate identification model to identify license plate numbers and license plate colors when the vehicle is identified to pass, and stores the identification result; (8) the number of cameras to be monitored is large, the system needs to run on a plurality of computers, and the main server needs to receive the detection result of the detection server; (9) in order to prevent the damage of the camera, the manager user can input the vehicle information directly and input the vehicle information into the system so as to calculate the time for entering the parking lot and calculate the parking cost in time;

(2) And (3) system architecture design:

the system application server is interacted with a user by using a browser and is responsible for fusing vehicle data, guard data, timing charging data and the like of enterprises, and data analysis and inquiry are provided for the user; the front end is developed by using technologies such as Vue frames or Bootstrap frames, and the back end is developed by using MyBatis, database technology MySQL, springBoot and Thymeleaf templates therein; the license plate recognition server is realized by using Python deep learning and is mainly responsible for reading flow from the monitoring server, then calling the vehicle detection and license plate character recognition model for recognition, and finally storing the recognition result into the vehicle information database.

Preferably, the coal enterprise management system is mainly designed by functions and architecture and database:

(1) And (3) main functional analysis:

the service scene of the vehicle information management system of the coal enterprise and the use requirement of the user on the product are combined, and the functional requirement of the system is analyzed and summarized by the method mainly comprises the following points: (1) providing basic data management functions for the administrator user by the system, including user management, role allocation and department management; (2) the system should have the system information monitoring functions such as log of user, operation log and the like; (3) the administrator user can manage the cameras according to actual needs, such as modifying camera information, deleting cameras and the like; (4) the user can preview the picture shot by the camera, set the detection area, check the detection state and whether to start detection; (5) the user can inquire specific information of the vehicle detected from the monitoring video according to the time, the camera name and other conditions, such as license plate color, number and snap-shot pictures; (6) the user can upload the vehicle information of the weighing data, edit and delete the vehicle information and check the importing result of the weighing data, and compare the vehicle information with the vehicle data detected by the system; (7) the system can acquire real-time video stream through the video stream address filled by the user, then carries out vehicle identification, calls a license plate identification model to identify license plate numbers and license plate colors when the vehicle is identified to pass, and stores the identification result; (8) the number of cameras to be monitored is large, the system needs to run on a plurality of computers, and the main server needs to receive the detection result of the detection server;

(2) And (3) system architecture design:

in terms of system architecture, the biggest difference between coal enterprise management and a conventional small and medium-sized parking lot management system is that a charging system is replaced by a coal weighing data recording system.

Preferably, the database design is an indispensable important part of the system development, and the good database design not only can reduce the storage space of data, but also can ensure the integrity of the data and improve the performance of the system; the database design mainly comprises a logic structure design and a physical structure design; the logic structure design mainly finds out the entity, attribute and relation among the entities in the system; as for the physical structure design, a physical structure most suitable for an application environment is selected according to a logic data model of a database;

through the demand analysis of the system, the system mainly comprises 6 entity structures of users, departments, roles, servers, license plate information and cameras; then analyzing the relationship of the 6 entities, wherein the relationship is obviously one-to-many between the user and the department, the relationship is also one-to-many between the user and the role, and the 6 entity structures of the user, the server, the license plate information and the camera are obvious; then analyzing the relationship of the 6 entities, wherein the relationship between the user and the department is obviously one-to-many, the relationship between the user and the role is also one-to-many, and the relationship between the user and the server as well as the relationship between the user and the camera are also one-to-many; the server and camera, camera and license plate information are also one-to-many relationships.

Compared with the prior art, the invention provides the vehicle management system with double systems and double modes based on the AI algorithm, which has the following beneficial effects:

1. the image processing method YOLOv4 in the conventional environment rapid recognition mode is an efficient and powerful target detection model, and can train an ultrafast and accurate target detector by using relatively moderate computing power equipment; during the training process of the detector, the influence of some most advanced research results on the target detector is verified; the SOTA method is improved, so that the SOTA method is more effective and more suitable for single GPU training;

2. the image processing method in the haze weather mode and coal vehicle management system uses the YOLOv3 recognition network algorithm based on the MobileNet V3 network structure, can ensure higher accuracy rate and certain instantaneity when recognizing the fuzzy license plate, can meet the requirements of real tasks, and has lower realization difficulty; the two algorithms are complementary to obtain a high-efficiency identification system, and the high-efficiency identification system can be selected according to the actual situation of the use scene, so that the advancement of the project is reflected;

3. the invention adopts a mode of complementation of various algorithms, and provides a double-mode double-system vehicle information management system, which is suitable for most application scenes in the whole country, meets most of requirements, adds a parking lot map into the parking lot management system, automatically records parking positions by users, combines the parking lot map, can more quickly find vehicles, improves user experience and parking lot use efficiency, solves the problems that most large-scale parking lots in the whole country are low in entering and exiting efficiency, users are easy to lose directions and other pain points in the parking lots, and in addition, solves the problems that vehicle recognition accuracy is low and speed is low when air visibility is low in a haze weather frequent region, solves the problem that coal enterprises are low in recognition accuracy of transporting coal truck license plates shielded by coal slime, meets the requirements of industrial production and common people on individuation, reduces unnecessary time waste, and becomes an effective tool for convenience people.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate and together with the embodiments of the invention and do not constitute a limitation to the invention, and in which:

FIG. 1 is a schematic diagram of a dual-system dual-mode vehicle management system based on AI algorithm;

FIG. 2 is an organizational chart of a VOC dataset file;

FIG. 3 is a flow chart of a model training and compression experiment;

FIG. 4 is a flow chart of a conventional environment quick recognition mode algorithm;

FIG. 5 is a haze weather pattern algorithm flow chart;

FIG. 6 is a schematic diagram of a license plate positioning process;

FIG. 7 is a schematic diagram of a general framework of a conventional small-to-medium parking lot management system;

FIG. 8 is a schematic diagram of a general framework of a coal enterprise management system;

fig. 9 is a database ER diagram.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

Referring to fig. 1-9, the present invention provides a technical solution: an AI algorithm-based vehicle management system with double systems and double modes is divided into the following parts: analyzing functional requirements and nonfunctional requirements of license plate recognition under different scene modes; each mode license plate image static identification task; comparing and selecting tasks by using an image processing algorithm of each mode; each mode license plate image dynamic identification task; selecting an image processing algorithm pruning compression task according to each mode; designing a vehicle information management system task; the vehicle detection model and the license plate recognition model are packaged and interact with the vehicle information management system, and the two-system and two-mode vehicle management system is divided into three major parts by combining the mathematical thinking based on analogy assumptions and the engineering thought designed from top to bottom as above: constructing a data set, training and compressing a model, designing and realizing a vehicle management system platform,

construction of a data set: the method has the advantages that various data sets with clear targets serving as license plate recognition are supplemented by using ImageNet, MSCOCO, PASCAVOC, CCPD and other public data sets and the principle of team self-shooting, and the Internet and the inside of the team have better image preprocessing processes for the data sets used by the method;

training and compression of the model: the identification of the vehicle identity adopts a vehicle identification technology, and the automatic identification of the license plate is a core function of the whole vehicle management system; different vehicle recognition algorithms are adopted for the vehicle recognition modes under different environments, wherein the vehicle recognition modes under different environments are as follows: a conventional environment rapid identification mode, a haze weather mode and a coal enterprise management system; the training of each license plate recognition model and the overall flow chart of the model pruning compression experiment are shown in figure 3;

design of a vehicle management system platform: the method is divided into main functions and architecture design of a conventional small and medium-sized parking lot management system, main functions and architecture design of a coal enterprise management system and database design.

In the construction of the data set, after enough pictures are collected, a labeling tool LabelImg can be used for labeling license plate pictures; wherein, chinese characters cannot be marked directly, none is used for replacing the Chinese characters, and Unicode codes are used for distinguishing the Chinese characters manually in the later period; the three modes comprise three different data sets, and taking a haze weather mode as an example, the haze weather mode data sets are constructed into a VOC data format; firstly, creating a file named Voc, and then creating Label_list. Txt, tranval. Txt, test. Txt and VOCdevkit file in the Voc file; wherein the Label_list file stores class labels of training data;

trainoval. Txt and test. Txt store pictures of training data and test data under the relative Voc file path and the corresponding paths of the labeling files; newly building a VOC2022 file in the VOCdevkit folder, and newly building a Annotations, imageSets, JPEGImage folder in the VOC 2022; the announcements folder stores license plates and license plate character labeling files; in Main files in ImageSets, the picture names of the test data and the training data are stored; the JPEGImages folder stores collected license plate pictures; stored in the Labels file is a markup file in YOLO format, as shown in fig. 2, which is an organizational chart of the VOC dataset file.

Conventional environment quick recognition mode: aiming at the key points and difficulties that the license plate detection and recognition algorithm needs to be realized in the conventional environment, the license plate detection and recognition algorithm can be divided into three modules: license plate detection, image processing and OCR license plate number recognition; the flow chart of the mode is shown in fig. 4;

(1) And (3) license plate detection: YOLOv4 target detection model based on improved SOTA method:

the yolov4 target detection algorithm consists of five basic components: CBM (conv+bn+mix activation function), CBL (conv+bn+leak_relu activation function), resunit (residual structure), CSPX (convolutional layer and Resunit module), SPP (pooling layer); the method has the main advantages that yolov4 is an efficient and powerful target detection model, and can train an ultrafast and accurate target detector by using relatively moderate computing power equipment; during the training process of the detector, the influence of some most advanced research results on the target detector is verified; the SOTA method is improved, so that the SOTA method is more effective and more suitable for single GPU training;

(2) Image processing: image noise reduction and binarization processing based on Opencv:

in order to further improve the recognition accuracy, the change of the light condition is that a certain difference exists in the detected and segmented image under normal conditions, so that the image is simply subjected to noise reduction and binarization processing by using an Opencv library, and the image is more suitable for OCR (optical character recognition) to be further recognized;

(3) OCR license plate number recognition: hundred-degree flying-oar OCR-based open-source text recognition:

the OCR is text recognition based on image detection, has the advantages of simple model training, small volume, high speed and high accuracy, and can quickly obtain license plate numbers through license plate images after recognition processing in a license plate recognition system, so that vehicles can quickly enter and exit a parking lot;

the image processing method YOLOv4 in the conventional environment rapid recognition mode is an efficient and powerful target detection model, and can train an ultrafast and accurate target detector by using relatively moderate computing power equipment; during the training process of the detector, the influence of some most advanced research results on the target detector is verified; the SOTA method is improved so that it is more efficient and more suitable for single GPU training.

Haze weather pattern: aiming at key points and difficulties which need to be realized by a license plate detection and recognition algorithm in haze weather, the license plate detection and recognition algorithm can be divided into three modules: defogging images, positioning license plates and recognizing license plate characters; the flow chart of the mode is shown in fig. 5;

(1) Image defogging: based on a prior defogging algorithm for fast guiding and filtering dark primary colors;

analyzing and comparing the decontamination effects of an image restoration defogging algorithm, a dark primary prior defogging algorithm and a fast guiding and filtering dark primary prior defogging algorithm based on an atmospheric scattering model, and selecting the defogging algorithm with the optimal effect based on the fast guiding and filtering dark primary prior defogging algorithm as a haze weather mode of the project;

(2) License plate positioning: license plate positioning algorithm based on edge detection;

edge detection is a common method in digital image processing, and the algorithm is to identify points with obvious brightness change in an image so as to realize edge contour distinction; in the license plate recognition system, the color difference between the color of the vehicle body and the color of the character is larger, so that the pixel lighting brightness jump at the edge of the license plate is large, and the outline of the edge is clear and is convenient to recognize; firstly, carrying out preprocessing such as denoising, gray stretching and the like on a license plate image to obtain an image with clear edge contour, counting the gray value of each pixel point in the image, and finding out a point with large gray value change, which is called a gray image jump point; the license plate position can be positioned according to the number of the jumping points, the specific flow of the algorithm is shown in fig. 6, the algorithm can rapidly and effectively identify the license plate of the vehicle in the actual application process, the realization difficulty is low, and the practical cost performance is extremely high;

(3) License plate character recognition: YOLOv3 identification network based on MobileNetV3 network structure;

training a haze weather license plate data set in a common target detection algorithm, and finding that a MobileNet V3 network structure-based Yolov3 recognition network algorithm compressed by a sensitivity-based method is selected to recognize a fuzzy license plate, so that higher accuracy can be ensured, a certain instantaneity is realized, and the requirements of real tasks can be met; the implementation difficulty of the recognition network algorithm is low, so that the algorithm is adopted as a core algorithm of the license plate character recognition process of the accurate mode of the haze weather of the project;

the image processing method in the haze weather mode and coal vehicle management system uses a YOLOv3 recognition network algorithm based on a MobileNet V3 network structure, so that the recognition of the fuzzy license plate can ensure higher accuracy and certain instantaneity, the requirements of real tasks can be met, and the realization difficulty of the recognition network algorithm is lower; the two algorithms are complementary to obtain an efficient identification system, and the efficient identification system can be selected according to the actual situation of the use scene as required, so that the advancement of the project is reflected.

The coal enterprise management system is an extension of the haze weather mode, a license plate character recognition algorithm trained by the same data set as the haze weather mode but different data sets is used (the algorithm has great advantage in the aspect of processing tiny shielding objects, and the influence of haze weather and coal slime on license plates has various commonalities, so that the effect is inferred to be feasible), and the difference is that the design and the realization of a vehicle management system platform are partially explained in detail.

The main functions and the architecture design of the conventional medium and small-sized parking lot management system are as follows:

(1) And (3) main functional analysis: the service scene of the conventional medium and small parking lot vehicle information management system and the use requirement of a user on products are combined, and the functional requirement of the system is analyzed and summarized by the service scene, which mainly comprises the following points:

(1) providing basic data management functions for the administrator user by the system, including user management, role allocation and department management; (2) providing a set of perfect automatic charging flow and charging information management system which can be controlled by a manager; (3) the system should have the system information monitoring functions such as log of user, operation log and the like; (4) the administrator user can manage the cameras according to actual needs, such as modifying camera information, deleting cameras and the like; (5) the user can preview the picture shot by the camera, set the detection area, check the detection state and whether to start detection; (6) the user can inquire specific information of the vehicle detected from the monitoring video according to the time, the camera name and other conditions, such as license plate color, number and snap-shot pictures; (7) the system can acquire real-time video stream through the video stream address filled by the user, then carries out vehicle identification, calls a license plate identification model to identify license plate numbers and license plate colors when the vehicle is identified to pass, and stores the identification result; (8) the number of cameras to be monitored is large, the system needs to run on a plurality of computers, and the main server needs to receive the detection result of the detection server; (9) in order to prevent the damage of the camera, the manager user can input the vehicle information directly and input the vehicle information into the system so as to calculate the time for entering the parking lot and calculate the parking cost in time;

(2) And (3) system architecture design:

the system application server is interacted with a user by using a browser and is responsible for fusing vehicle data, guard data, timing charging data and the like of enterprises, and data analysis and inquiry are provided for the user; the front end is developed by using technologies such as Vue frames or Bootstrap frames, and the back end is developed by using MyBatis, database technology MySQL, springBoot and Thymeleaf templates therein; the license plate recognition server is realized by using Python deep learning and is mainly responsible for reading flow from the monitoring server, then calling the vehicle detection and license plate character recognition model for recognition, and finally storing the recognition result into the vehicle information database; the general frame is shown in fig. 7.

Major functions and architecture design and database design of the coal enterprise management system:

(1) And (3) main functional analysis:

the service scene of the vehicle information management system of the coal enterprise and the use requirement of the user on the product are combined, and the functional requirement of the system is analyzed and summarized by the method mainly comprises the following points: (1) providing basic data management functions for the administrator user by the system, including user management, role allocation and department management; (2) the system should have the system information monitoring functions such as log of user, operation log and the like; (3) the administrator user can manage the cameras according to actual needs, such as modifying camera information, deleting cameras and the like; (4) the user can preview the picture shot by the camera, set the detection area, check the detection state and whether to start detection; (5) the user can inquire specific information of the vehicle detected from the monitoring video according to the time, the camera name and other conditions, such as license plate color, number and snap-shot pictures; (6) the user can upload the vehicle information of the weighing data, edit and delete the vehicle information and check the importing result of the weighing data, and compare the vehicle information with the vehicle data detected by the system; (7) the system can acquire real-time video stream through the video stream address filled by the user, then carries out vehicle identification, calls a license plate identification model to identify license plate numbers and license plate colors when the vehicle is identified to pass, and stores the identification result; (8) the number of cameras to be monitored is large, the system needs to run on a plurality of computers, and the main server needs to receive the detection result of the detection server;

(2) And (3) system architecture design:

in terms of system architecture, the biggest difference between coal enterprise management and a conventional small and medium-sized parking lot management system is that a charging system is replaced by a coal weighing data recording system, and the general framework is shown in fig. 8.

The database design is an indispensable important part of the system development, and the good database design not only can reduce the storage space of data, but also can ensure the integrity of the data and improve the performance of the system; the database design mainly comprises a logic structure design and a physical structure design; the logic structure design mainly finds out the entity, attribute and relation among the entities in the system; as for the physical structure design, a physical structure most suitable for an application environment is selected according to a logic data model of a database;

through the demand analysis of the system, the system mainly comprises 6 entity structures of users, departments, roles, servers, license plate information and cameras; then analyzing the relationship of the 6 entities, wherein the relationship is obviously one-to-many between the user and the department, the relationship is also one-to-many between the user and the role, and the 6 entity structures of the user, the server, the license plate information and the camera are obvious; then analyzing the relationship of the 6 entities, wherein the relationship between the user and the department is obviously one-to-many, the relationship between the user and the role is also one-to-many, and the relationship between the user and the server as well as the relationship between the user and the camera are also one-to-many; the server and camera, camera and license plate information are also one-to-many, and the ER diagram is shown in FIG. 9.

Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.

Claims (8)

1. The utility model provides a vehicle management system with dual system dual mode based on AI algorithm, includes the construction of dataset, the training of model and the construction of compression and the design of vehicle management system platform, its characterized in that:

construction of a data set: the method has the advantages that various data sets with clear targets serving as license plate recognition are supplemented by using ImageNet, MSCOCO, PASCAVOC, CCPD and other public data sets and the principle of team self-shooting, and the Internet and the inside of the team have better image preprocessing processes for the data sets used by the method;

training and compression of the model: the identification of the vehicle identity adopts a vehicle identification technology, and the automatic identification of the license plate is a core function of the whole vehicle management system; different vehicle recognition algorithms are adopted for the vehicle recognition modes under different environments, wherein the vehicle recognition modes under different environments are as follows: a conventional environment rapid identification mode, a haze weather mode and a coal enterprise management system;

design of a vehicle management system platform: the method is divided into main functions and architecture design of a conventional small and medium-sized parking lot management system, main functions and architecture design of a coal enterprise management system and database design.

2. The AI algorithm-based vehicle management system with dual systems and modes of claim 1, wherein: in the construction of the data set, after enough pictures are collected, a labeling tool LabelImg can be used for labeling license plate pictures; wherein, chinese characters cannot be marked directly, none is used for replacing the Chinese characters, and Unicode codes are used for distinguishing the Chinese characters manually in the later period; the three modes comprise three different data sets, and taking a haze weather mode as an example, the haze weather mode data sets are constructed into a VOC data format; firstly, creating a file named Voc, and then creating Label_list. Txt, tranval. Txt, test. Txt and VOCdevkit file in the Voc file; wherein the Label_list file stores class labels of training data; trainoval. Txt and test. Txt store pictures of training data and test data under the relative Voc file path and the corresponding paths of the labeling files; newly building a VOC2022 file in the VOCdevkit folder, and newly building a Annotations, imageSets, JPEGImage folder in the VOC 2022; the announcements folder stores license plates and license plate character labeling files; in Main files in ImageSets, the picture names of the test data and the training data are stored; the JPEGImages folder stores collected license plate pictures; the Labels file stores a marking file in a Yolo format and a VOC data set file organization chart.

3. The AI algorithm-based vehicle management system with dual systems and modes of claim 1, wherein: conventional environment quick recognition mode: aiming at the key points and difficulties that the license plate detection and recognition algorithm needs to be realized in the conventional environment, the license plate detection and recognition algorithm can be divided into three modules: license plate detection, image processing and OCR license plate number recognition;

(1) And (3) license plate detection: YOLOv4 target detection model based on improved SOTA method:

the yolov4 target detection algorithm consists of five basic components: CBM (conv+bn+mix activation function), CBL (conv+bn+leak_relu activation function), resunit (residual structure), CSPX (convolutional layer and Resunit module), SPP (pooling layer); the method has the main advantages that yolov4 is an efficient and powerful target detection model, and can train an ultrafast and accurate target detector by using relatively moderate computing power equipment; during the training process of the detector, the influence of some most advanced research results on the target detector is verified; the SOTA method is improved, so that the SOTA method is more effective and more suitable for single GPU training;

(2) Image processing: image noise reduction and binarization processing based on Opencv:

in order to further improve the recognition accuracy, the change of the light condition is that a certain difference exists in the detected and segmented image under normal conditions, so that the image is simply subjected to noise reduction and binarization processing by using an Opencv library, and the image is more suitable for OCR (optical character recognition) to be further recognized;

(3) OCR license plate number recognition: hundred-degree flying-oar OCR-based open-source text recognition:

the OCR is text recognition based on image detection, has the advantages of simple model training, small volume, high speed and high accuracy, and can quickly obtain license plate numbers through recognition processed license plate images in a license plate recognition system, thereby realizing quick vehicle entering and exiting of parking lots.

4. The AI algorithm-based vehicle management system with dual systems and modes of claim 1, wherein: haze weather pattern: aiming at key points and difficulties which need to be realized by a license plate detection and recognition algorithm in haze weather, the license plate detection and recognition algorithm can be divided into three modules: defogging images, positioning license plates and recognizing license plate characters;

(1) Image defogging: based on a prior defogging algorithm for fast guiding and filtering dark primary colors;

analyzing and comparing the decontamination effects of an image restoration defogging algorithm, a dark primary prior defogging algorithm and a fast guiding and filtering dark primary prior defogging algorithm based on an atmospheric scattering model, and selecting the defogging algorithm with the optimal effect based on the fast guiding and filtering dark primary prior defogging algorithm as a haze weather mode of the project;

(2) License plate positioning: license plate positioning algorithm based on edge detection;

edge detection is a common method in digital image processing, and the algorithm is to identify points with obvious brightness change in an image so as to realize edge contour distinction; in the license plate recognition system, the color difference between the color of the vehicle body and the color of the character is larger, so that the pixel lighting brightness jump at the edge of the license plate is large, and the outline of the edge is clear and is convenient to recognize; firstly, carrying out preprocessing such as denoising, gray stretching and the like on a license plate image to obtain an image with clear edge contour, counting the gray value of each pixel point in the image, and finding out a point with large gray value change, which is called a gray image jump point; the license plate position can be positioned according to the number of the jump points, the algorithm can quickly and effectively identify the license plate of the vehicle in the actual application process, the realization difficulty is low, and the practical cost performance is extremely high;

(3) License plate character recognition: YOLOv3 identification network based on MobileNetV3 network structure;

training a haze weather license plate data set in a common target detection algorithm, and finding that a MobileNet V3 network structure-based Yolov3 recognition network algorithm compressed by a sensitivity-based method is selected to recognize a fuzzy license plate, so that higher accuracy can be ensured, a certain instantaneity is realized, and the requirements of real tasks can be met; and the realization difficulty of the recognition network algorithm is low, so the algorithm is adopted as a core algorithm of the license plate character recognition process of the accurate mode of the haze weather of the project.

5. The AI algorithm-based vehicle management system with dual systems and modes of claim 1, wherein: the coal enterprise management system is an extension of the haze weather mode, a license plate character recognition algorithm trained by the same data set as the haze weather mode but different data sets is used (the algorithm has great advantage in the aspect of processing tiny shielding objects, and the influence of haze weather and coal slime on license plates has various commonalities, so that the effect is inferred to be feasible), and the difference is that the design and the realization of a vehicle management system platform are partially explained in detail.

6. The AI algorithm-based vehicle management system with dual systems and modes of claim 1, wherein: the main functions and the architecture design of the conventional medium and small-sized parking lot management system are as follows:

(1) And (3) main functional analysis: the service scene of the conventional medium and small parking lot vehicle information management system and the use requirement of a user on products are combined, and the functional requirement of the system is analyzed and summarized by the service scene, which mainly comprises the following points:

(1) providing basic data management functions for the administrator user by the system, including user management, role allocation and department management; (2) providing a set of perfect automatic charging flow and charging information management system which can be controlled by a manager; (3) the system should have the system information monitoring functions such as log of user, operation log and the like; (4) the administrator user can manage the cameras according to actual needs, such as modifying camera information, deleting cameras and the like; (5) the user can preview the picture shot by the camera, set the detection area, check the detection state and whether to start detection; (6) the user can inquire specific information of the vehicle detected from the monitoring video according to the time, the camera name and other conditions, such as license plate color, number and snap-shot pictures; (7) the system can acquire real-time video stream through the video stream address filled by the user, then carries out vehicle identification, calls a license plate identification model to identify license plate numbers and license plate colors when the vehicle is identified to pass, and stores the identification result; (8) the number of cameras to be monitored is large, the system needs to run on a plurality of computers, and the main server needs to receive the detection result of the detection server; (9) in order to prevent the damage of the camera, the manager user can input the vehicle information directly and input the vehicle information into the system so as to calculate the time for entering the parking lot and calculate the parking cost in time;

(2) And (3) system architecture design:

the system application server is interacted with a user by using a browser and is responsible for fusing vehicle data, guard data, timing charging data and the like of enterprises, and data analysis and inquiry are provided for the user; the front end is developed by using technologies such as Vue frames or Bootstrap frames, and the back end is developed by using MyBatis, database technology MySQL, springBoot and Thymeleaf templates therein; the license plate recognition server is realized by using Python deep learning and is mainly responsible for reading flow from the monitoring server, then calling the vehicle detection and license plate character recognition model for recognition, and finally storing the recognition result into the vehicle information database.

7. The AI algorithm-based vehicle management system with dual systems and modes of claim 1, wherein: major functions and architecture design and database design of the coal enterprise management system:

(1) And (3) main functional analysis:

the service scene of the vehicle information management system of the coal enterprise and the use requirement of the user on the product are combined, and the functional requirement of the system is analyzed and summarized by the method mainly comprises the following points: (1) providing basic data management functions for the administrator user by the system, including user management, role allocation and department management; (2) the system should have the system information monitoring functions such as log of user, operation log and the like; (3) the administrator user can manage the cameras according to actual needs, such as modifying camera information, deleting cameras and the like; (4) the user can preview the picture shot by the camera, set the detection area, check the detection state and whether to start detection; (5) the user can inquire specific information of the vehicle detected from the monitoring video according to the time, the camera name and other conditions, such as license plate color, number and snap-shot pictures; (6) the user can upload the vehicle information of the weighing data, edit and delete the vehicle information and check the importing result of the weighing data, and compare the vehicle information with the vehicle data detected by the system; (7) the system can acquire real-time video stream through the video stream address filled by the user, then carries out vehicle identification, calls a license plate identification model to identify license plate numbers and license plate colors when the vehicle is identified to pass, and stores the identification result; (8) the number of cameras to be monitored is large, the system needs to run on a plurality of computers, and the main server needs to receive the detection result of the detection server;

(2) And (3) system architecture design:

in terms of system architecture, the biggest difference between coal enterprise management and a conventional small and medium-sized parking lot management system is that a charging system is replaced by a coal weighing data recording system.

8. The AI algorithm-based vehicle management system with dual systems and modes of claim 1, wherein: the database design is an indispensable important part of the system development, and the good database design not only can reduce the storage space of data, but also can ensure the integrity of the data and improve the performance of the system; the database design mainly comprises a logic structure design and a physical structure design; the logic structure design mainly finds out the entity, attribute and relation among the entities in the system; as for the physical structure design, a physical structure most suitable for an application environment is selected according to a logic data model of a database;

through the demand analysis of the system, the system mainly comprises 6 entity structures of users, departments, roles, servers, license plate information and cameras; then analyzing the relationship of the 6 entities, wherein the relationship is obviously one-to-many between the user and the department, the relationship is also one-to-many between the user and the role, and the 6 entity structures of the user, the server, the license plate information and the camera are obvious; then analyzing the relationship of the 6 entities, wherein the relationship between the user and the department is obviously one-to-many, the relationship between the user and the role is also one-to-many, and the relationship between the user and the server as well as the relationship between the user and the camera are also one-to-many; the server and camera, camera and license plate information are also one-to-many relationships.

Images