Method and device for recognizing license plate information of covered vehicles based on vision technology
The method enhances license plate recognition for covered vehicles by integrating ETC system data and edge computing for deep restoration and error correction, addressing challenges of occlusion and poor image quality for improved accuracy and efficiency.
Patent Information
- Application Number
- CN202311608795.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-11-29
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2043-11-29
AI Technical Summary
Conventional methods for recognizing license plate information of covered vehicles face challenges due to occlusion, deformation, and poor image quality, leading to inaccurate identification.
A method combining license plate image acquisition with on-board ETC system data, utilizing edge computing for deep restoration and error correction, and employing multiple cameras and encryption protocols to enhance recognition accuracy and security.
Improves the reliability and efficiency of license plate recognition for covered vehicles by correcting errors and restoring obscured or damaged images in real-time, ensuring accurate vehicle identification.
Smart Images

Figure CN117710952B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of machine vision technology, and in particular to a method and device for recognizing license plate information of a masked vehicle based on vision technology. Background Art
[0002] A covered vehicle (also called a camouflaged vehicle or concealed vehicle) generally refers to a vehicle whose license plate is blocked, disguised or hidden to prevent identification. The goal of license plate information recognition methods is to identify the license plate information of these covered vehicles as accurately as possible. Conventional methods often use image processing and computer vision techniques, such as character recognition (OCR) to identify license plates. Covered vehicles use a variety of camouflage methods, such as occlusion, projection, and deformation, which can make it difficult for traditional OCR technology to accurately identify. Summary of the invention
[0003] In order to solve the above technical problems, the present invention proposes a method and device for recognizing license plate information of covered vehicles based on vision technology to solve at least one of the above technical problems.
[0004] The present application provides a method for recognizing license plate information of a covered vehicle based on visual technology, comprising the following steps:
[0005] Step S1: Capturing the license plate image through a preset license plate acquisition camera, thereby obtaining license plate image data;
[0006] Step S2: Collect the device number of the vehicle-mounted ETC system through the intelligent roadside equipment installed on the roadside, so as to obtain the device number data;
[0007] Step S3: The license plate image data and the device number data are processed and stored by the edge computing unit, thereby obtaining the license plate device storage data;
[0008] Step S4: Obtain the number data of the vehicle-mounted equipment to be detected and the license plate image data to be detected through the intelligent roadside equipment, and use the license plate device storage data and the number data of the vehicle-mounted equipment to be detected to deeply restore the license plate image data to be detected through the edge computing unit, so as to obtain the license plate data of the masked vehicle.
[0009] In the present invention, by combining license plate image acquisition and on-board ETC system data acquisition, the method can improve the recognition accuracy of covered vehicles. Traditional license plate recognition systems are interfered by factors such as license plate stains, occlusions, and backlighting, while this method can improve the reliability of recognition with the assistance of ETC system data. Through the deep restoration technology in step S4, even if the license plate image quality is poor or there is occlusion, the license plate data can be restored. This helps to solve problems that are difficult to handle with visual analysis, such as covered or damaged license plates. Since the license plate image acquisition and equipment number acquisition in the method are carried out simultaneously, real-time or efficient recognition can be achieved without the need for additional parking waiting time. This helps to improve the efficiency of road traffic management. By associating the license plate image data with the on-board ETC system data, it can be ensured that the license plate information is associated with the equipment number of a specific vehicle.
[0010] Preferably, the license plate image data includes first license plate image data and second license plate image data, the license plate acquisition camera includes a first license plate acquisition camera and a second license plate acquisition camera, and step S1 specifically includes:
[0011] Controlling the first vehicle acquisition camera to acquire first vehicle image data at a first vehicle acquisition angle, thereby acquiring first vehicle image data;
[0012] The second vehicle acquisition camera is controlled to acquire the second vehicle image data at a second vehicle acquisition angle, thereby acquiring the second vehicle image data.
[0013] In the present invention, by using a first license plate acquisition camera and a second license plate acquisition camera, image data of two vehicles are acquired at the same time, or license plate data of vehicles are acquired from different angles, thereby reducing the error problem caused by illumination or angle, and improving the efficiency of data acquisition, especially in high-flow traffic scenes, license plate images can be acquired more quickly. Since the images of two vehicles are acquired at the same time, the influence of vehicle interference in the traffic flow on data acquisition is reduced. Even if multiple vehicles pass by at the same time, their license plate image data can be acquired effectively. For traffic monitoring and management, especially in scenes where multiple vehicles need to be identified in real time, acquiring image data of multiple vehicles at the same time can help improve efficiency.
[0014] Preferably, step S2 specifically includes:
[0015] Step S21: Communicate with the vehicle-mounted ETC system device through the intelligent roadside device installed on the roadside, thereby obtaining device communication data;
[0016] Step S22: Establish a connection between the intelligent roadside device and the vehicle-mounted ETC system device according to the device communication data, and collect the vehicle-mounted ETC system device number through the intelligent roadside device to obtain the primary device number data;
[0017] Step S23: performing error detection and correction on the primary device number data, thereby obtaining revised device number data;
[0018] Step S24: storing the corrected device number data, thereby obtaining the device number data.
[0019] The present invention utilizes the communication data of the on-board ETC system, establishes a connection with the communication of the ETC system, can obtain the OBU device number associated with a specific vehicle, associates the license plate information with the OBU device number of the vehicle, and ensures the consistency and accuracy of the data. In step S23, the primary device number data is error detected and corrected, which helps to ensure the accuracy of the device number, can improve the accurate identification of the vehicle, and reduce the misidentification caused by the device number error. Step S24 stores the corrected OBU data to form the license plate device storage data, so that the system can quickly access and retrieve the OBU data associated with the vehicle, thereby improving efficiency in the license plate information recognition process. By establishing a connection with the OBU device and correcting the device number data, the reliability of vehicle identification can be improved, which is very important for ensuring that the correct license plate information is associated with a specific vehicle.
[0020] Preferably, in step S21, the communication is encrypted by using an intelligent roadside device authentication encryption value calculation formula, wherein the intelligent roadside device authentication encryption value calculation formula is specifically:
[0021] ;
[0022] Authenticate encrypted values for intelligent roadside equipment, is the communication data quantity data, is the communication data index item, For the Intelligent roadside equipment communication data, is the communication time range data, is the communication data change rate, is the communication time data, is the communication propagation angle data, To communicate relative angle data, Load data for intelligent roadside equipment, Encode data for intelligent roadside equipment data.
[0023] The present invention constructs a calculation formula for the authentication encryption value of intelligent roadside equipment, which provides stable and reliable data support in intelligent Internet of Things, especially in the encryption communication calculation of intelligent roadside equipment. It is the encrypted value for intelligent roadside equipment authentication, representing the final encrypted value for authentication. For the The communication data of intelligent roadside equipment represents a part of the communication data. Communication propagation angle data, indicating the direction of the communication device or the propagation direction of the signal. For example, if a device needs to adjust its antenna direction to communicate with another device, Indicates the direction angle of the antenna. The value is usually expressed in degrees or radians, depending on the specific standard used. Communication relative angle data, used to indicate the direction or angle related to communication. The specific meaning of also depends on the specific application scenario. In communication, Indicates the incident angle of the communication signal, or the relative direction between devices. For example, if two vehicles need to communicate with each other, Indicates the relative direction of one vehicle to another. The value of is usually expressed in degrees or radians. The present invention can improve the security of communications. This helps reduce the risk of unauthorized access and data leakage. The calculation formula includes consideration of the rate of change of the communication data, which can be used to detect inconsistencies or tampering attempts of the data. If the data changes or is tampered with, the authentication will fail, thereby protecting the integrity of the data. The parameters in the formula and The load and data encoding characteristics of the communication device are taken into account, and the performance and characteristics of the device are taken into account during the authentication process, so as to better adapt to different types of devices. The invention aims to improve the security and reliability of intelligent roadside device authentication, while taking into account the characteristics of the communication device, thereby enhancing the performance and security of the entire communication system. This is essential to ensure secure communication between intelligent roadside devices and the integrity of data, especially in scenarios involving vehicle communications and intelligent transportation systems.
[0024] Preferably, step S23 is specifically:
[0025] Step S231: creating an original device number according to the primary device number data, thereby obtaining the original device number data;
[0026] Step S232: creating an error detection area and performing data replication according to the original device number data, thereby obtaining the error detection device number data;
[0027] Step S233: segmenting the error detection device number data to obtain device number segmentation data;
[0028] Step S234: extracting character features and pixel density features from the device number segmentation data, thereby obtaining device number character feature data and device number pixel density feature data;
[0029] Step S235: performing a first recognition calculation on the device number character feature data and the device number pixel density feature data, thereby obtaining device number error detection position data;
[0030] Step S236: performing shape edge curvature feature extraction and projection density feature extraction on the device number segmentation data, thereby obtaining shape edge curvature feature data and projection density feature data;
[0031] Step S237: performing a second identification calculation on the shape edge curvature feature data and the projection density feature data, thereby obtaining device number error detection content data;
[0032] Step S238: performing correction processing on the device number segmentation data according to the device number error detection position data and the device number error detection content data, thereby obtaining corrected device number data;
[0033] Step S239: Perform character combination on the corrected device number data to obtain the corrected device number data.
[0034] In the present invention, step S231 converts the device number data from the primary state to the original state by creating the original device number, which is used for comparison with subsequent error detection and correction. Step S232 creates an error detection area for determining the part containing the error and copies the data for later use, which helps to narrow the scope of the error and reduce the computational complexity of error detection. Step S233 divides the device number into smaller parts so that each part can be individually error detected and corrected. This improves the accuracy and reliability of the system. Step S234 extracts character features and pixel density features from the device number segmentation data, providing information about the shape and density of the character, which helps with subsequent error detection. Step S235 uses the character features and pixel density features to perform a first recognition calculation to determine the location of the error detection, which helps to determine the location where the error exists in the device number. Step S236 extracts the shape edge curvature features and projection density features, which can provide more information about the character shape for further error detection. Step S237 uses the shape edge curvature features and projection density features to perform a second recognition calculation to determine the content of the error detection, which helps to determine the specific character or content of the error. Step S238 corrects the device number segmentation data according to the position and content of the error detection, and combines the corrected partial characters into the revised OBU data, thereby ensuring the accuracy of the device number.
[0035] Preferably, step S3 specifically includes:
[0036] Step S31: The license plate image data and the device number data are transmitted to the edge computing unit, and the edge computing unit performs parsing to obtain the license plate image parsing data and the device number parsing data;
[0037] Step S32: performing intelligent image correction processing on the license plate image analysis data using preset license plate image standard parameters, thereby obtaining license plate image correction data;
[0038] Step S33: Associating and merging the license plate image correction data and the device number parsing data, thereby obtaining license plate device number associated merged data;
[0039] Step S34: the license plate device number associated merged data is access-controlled marked and stored through a preset multi-level data access control strategy, thereby obtaining the license plate device storage data.
[0040] In the present invention, step S31 is responsible for transmitting the license plate image data and the equipment number data to the edge computing unit, and parsing the data, which helps to associate the two and ensure that the subsequent steps can correctly process the data. Step S32 performs intelligent image correction processing on the license plate image parsing data to correct the license plate image to a standardized format according to the preset license plate image standard parameters, which helps to improve the accuracy and readability of subsequent license plate information. Step S33 associates and merges the license plate image correction data with the equipment number parsing data to ensure that the license plate information is correctly matched with the vehicle's equipment number, providing accurate data for subsequent access control and storage. Step S34 uses a multi-level data access control strategy to perform access control annotation and storage on the license plate equipment number associated merge data, ensuring that only authorized users can access and use the data, thereby improving data security and privacy protection.
[0041] Preferably, the intelligent image correction processing in step S32 is performed by using an intelligent image correction processing calculation formula, wherein the intelligent image correction processing calculation formula is specifically:
[0042] ;
[0043] Correction data for license plate images, is the lateral blur corrected for the license plate image, is the longitudinal blur correction of the license plate image, is the horizontal coordinate point of the license plate image, is the horizontal coordinate point of the center point of the license plate image, is the longitudinal coordinate point of the license plate image, is the longitudinal coordinate point of the center point of the license plate image, To correct the fluctuation period term, is the distance correction factor, is the original license plate image intensity distribution data.
[0044] The present invention constructs an intelligent image correction processing calculation formula, which corrects the license plate image to improve its clarity and readability. is the horizontal coordinate point of the license plate image and It is the longitudinal coordinate point of the license plate image, indicating the coordinate point on the image, and is used to specify the corrected pixel position. The lateral blur corrected for the license plate image and The longitudinal blurriness corrected for the license plate image represents the standard deviation of the image, which is used to control the lateral and longitudinal blurriness of the image correction. is the horizontal coordinate point of the center point of the license plate image and The longitudinal coordinate point of the center point of the license plate image represents the coordinates of the center point of the image and is used to specify the center position of the correction. The correction fluctuation period term represents the wavelength of the light wave and is used to control the fluctuation period in the correction. It is the correction distance factor, which represents the distance on the image and is used to specify the correction distance factor. is the intensity distribution data of the original license plate image, which represents the intensity distribution of the original license plate image. This formula describes the process of intelligent image correction. To control the blur of the image, the cosine function To introduce the wave effect, as well as the image intensity distribution to represent the characteristics of the original license plate image. The correction term in the formula can correct the distortion and blur in the license plate image and improve the clarity and quality of the license plate image. This helps to recognize the license plate information more accurately. By adjusting parameters such as the horizontal and vertical blur ( and )、Fluctuation period term( ) and the correction distance factor ( ), the method can adapt to different lighting conditions, angles and distances, thereby improving the license plate recognition capability in various environments. By using the cosine function to introduce the wave effect, the misrecognition caused by the reflection, distortion or occlusion of the license plate image can be reduced, and the stability of the license plate recognition can be improved. The present invention improves the accuracy and stability of license plate information recognition.
[0045] Preferably, step S4 is specifically:
[0046] Step S41: Acquire the number data of the vehicle-mounted device to be detected and the image data of the license plate to be detected through the intelligent roadside device;
[0047] Step S42: verifying and correcting the vehicle-mounted device number data through the edge computing unit, thereby obtaining error correction data of the vehicle-mounted device number to be detected;
[0048] Step S43: matching the data stored in the license plate device with the error correction data of the vehicle-mounted device number to obtain matching data of the license plate to be detected;
[0049] Step S44: performing a license plate data integrity test on the license plate image data to be detected, thereby obtaining license plate data integrity test data, thereby obtaining license plate data integrity test data, wherein the license plate data integrity test data includes complete license plate data and missing license plate data;
[0050] Step S45: if the license plate data integrity detection data is determined to be complete license plate data, the license plate image data to be detected is determined to be the license plate data of the masked vehicle;
[0051] Step S46: determining that the license plate data integrity detection data is license plate missing data, then performing image mapping on the license plate image data to be detected according to the license plate image data in the license plate device storage data, thereby obtaining image mapping data;
[0052] Step S47: performing confidence calculation on the image mapping data, thereby obtaining image mapping confidence data;
[0053] Step S48: when it is determined that the image mapping confidence data is the first image mapping confidence data, the license plate image data in the license plate device storage data is determined as the masked vehicle license plate data;
[0054] Step S49: When it is determined that the image mapping confidence data is the second image mapping confidence data, the license plate image data to be detected is deeply restored to obtain the license plate data of the masked vehicle.
[0055] In the present invention, step S41 is responsible for obtaining the vehicle-mounted device number data to be detected and the license plate image data to be detected. Step S42 ensures the accuracy of the device number by checking and correcting the vehicle-mounted device number data to be detected, thereby improving the reliability of vehicle identification. Step S43 matches the vehicle-mounted device number correction data with the license plate device storage data according to the vehicle-mounted device number correction data to be detected, thereby ensuring the correct correspondence between the device number and the license plate information. Step S44 performs integrity detection on the license plate image data to be detected to determine whether the license plate is complete and distinguish between masked vehicles and other situations, such as partial occlusion or damage of the license plate. Step S46 performs image mapping on the license plate image data to be detected, and step S47 calculates the confidence of the image mapping, determines the credibility of the image mapping, and then determines whether deep restoration is needed. Step S49 performs deep restoration on the license plate image data to be detected to obtain the license plate data of the masked vehicle, which plays a key role when the license plate data is missing or incomplete. The present invention performs multi-level identification and verification on the license plate information of the masked vehicle, thereby improving the accuracy and credibility of vehicle identification.
[0056] Preferably, the deep restoration in step S49 is performed through an image restoration process using a vehicle license plate deep restoration image model, wherein the steps of constructing the vehicle license plate deep restoration image model are specifically as follows:
[0057] Step S491: Acquire standard covered vehicle license plate image data and corresponding standard covered vehicle license plate data;
[0058] Step S492: extracting features from the standard covered-plate vehicle license plate image data, thereby obtaining the standard covered-plate vehicle license plate image feature data;
[0059] Step S493: performing complexity processing on the image feature data of the standard covered vehicle license plate, thereby obtaining image feature complexity data;
[0060] Step S494: generating image feature convolution layer data according to the image feature complexity data, and performing convolution calculation on the standard masked vehicle license plate image feature data using the image feature convolution layer data, thereby obtaining the standard masked vehicle license plate image convolution layer data;
[0061] Step S495: performing pooling layer calculation on the convolution layer data of the standard masked vehicle license plate image, thereby obtaining the pooling layer data of the standard masked vehicle license plate image;
[0062] Step S496: performing full connection calculation on the pooling layer data of the standard masked vehicle license plate image, thereby obtaining full connection weight data of the standard masked vehicle license plate image;
[0063] Step S497: Annotate the fully connected weighted data of the standard masked vehicle license plate image according to the standard masked vehicle license plate data, thereby obtaining a vehicle license plate deep restoration image model.
[0064] The model of the present invention can be used to restore the license plate image of the covered vehicle, and the license plate information can be effectively restored even in the case of blur, noise or damage. This helps to improve the accuracy of vehicle recognition, especially in security monitoring and traffic management. The model uses deep learning technology, especially convolutional neural network (CNN) and fully connected network, to extract features and perform complex calculations on the image, and can learn higher-level image features, thereby improving the restoration effect. The license plate image is affected by different lighting, weather conditions and angles, so higher-level features are required to identify the license plate in different situations. The introduction of image feature convolution layer data can improve the model's adaptability to these complex scenes and reduce the probability of recognition errors. The convolution calculation in this step can effectively fuse the image features with the standard covered vehicle license plate image feature data. This fusion helps to improve the generalization performance of the deep learning model, so that it can process license plate images of different types and styles, not just for specific situations. By using standard covered vehicle license plate image data and corresponding standard covered vehicle license plate data, the model can be effectively trained and labeled. This ensures that the model can accurately restore the license plate image during the restoration process, especially in terms of license plate image diversity.
[0065] Preferably, a device for recognizing license plate information of a covered vehicle based on visual technology is used to perform the method for recognizing license plate information of a covered vehicle based on visual technology as described above, and the device for recognizing license plate information of a covered vehicle based on visual technology comprises:
[0066] The license plate image acquisition module is used to acquire the license plate image through a preset license plate acquisition camera, thereby obtaining the license plate image data;
[0067] The equipment number data collection module is used to collect the equipment number of the vehicle-mounted ETC system through the intelligent roadside equipment installed on the roadside, so as to obtain the equipment number data;
[0068] A license plate device number storage module is used to process and store the license plate image data and device number data through the edge computing unit, thereby obtaining the license plate device storage data;
[0069] The retrieval and deep restoration module is used to obtain the numbering data of the on-board equipment to be detected through the intelligent roadside equipment, and use the data stored in the license plate device to retrieve and deeply restore the numbering data of the on-board equipment to be detected through the edge computing unit, so as to obtain the license plate data of the masked vehicle.
[0070] The beneficial effect of the present invention is that the method can comprehensively collect license plate image data and vehicle ETC system equipment number data, thereby ensuring the multi-dimensional information collection of the vehicle and providing sufficient data support for subsequent identification and verification. The present invention includes the collection and verification process of the equipment number, ensuring the accuracy of the equipment number, ensuring that the vehicle and its corresponding OBU equipment are correctly matched, and avoiding misidentification. The license plate image data and equipment number data are transmitted to the edge computing unit and stored as license plate device storage data. The multi-level data storage adopted allows the data to be effectively managed for subsequent query and identification. The deep restoration technology in the present invention is applied to the license plate image data of the vehicle to be detected, and the license plate information can still be restored when the license plate image is blocked, defaced, or backlit. Through the deep restoration technology, the method can highly reliably identify the license plate information of the masked vehicle, which is of great significance in improving traffic management, safety monitoring, and crime prevention and control, and helps to reduce illegal acts and improve the level of public security. The edge computing unit allows high efficiency in real time and can process more vehicles and data, so as to cope with peak traffic flow and the growing number of vehicles. BRIEF DESCRIPTION OF THE DRAWINGS
[0071] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting implementations made with reference to the following drawings:
[0072] Figure 1 A flowchart showing a method for recognizing license plate information of a masked vehicle based on visual technology according to an embodiment of the present invention;
[0073] Figure 2 A flowchart of step S2 of an embodiment is shown;
[0074] Figure 3 A flowchart of step S23 of an embodiment is shown;
[0075] Figure 4 A flowchart of step S3 of an embodiment is shown;
[0076] Figure 5 A flowchart of step S4 of an embodiment is shown;
[0077] Figure 6 A step flow chart of step S49 of an embodiment is shown. DETAILED DESCRIPTION
[0078] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.
[0079] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.
[0080] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.
[0081] See also Figures 1 to 6 The present application provides a method for recognizing license plate information of a masked vehicle based on visual technology, comprising the following steps:
[0082] Step S1: Capturing the license plate image through a preset license plate acquisition camera, thereby obtaining license plate image data;
[0083] Specifically, for example, license plate acquisition cameras use advanced image acquisition technology, such as high-resolution CMOS sensors and lens components, to acquire license plate image data. These cameras are specifically industrial cameras, such as Basler ace cameras.
[0084] Step S2: Collect the device number of the vehicle-mounted ETC system through the intelligent roadside equipment installed on the roadside, so as to obtain the device number data;
[0085] Specifically, for example, the intelligent roadside equipment uses radio frequency identification (RFID) or other communication technologies to communicate with the vehicle-mounted ETC system OBU device to obtain the OBU device number data. For example, the RSU can use 5.8 GHz radio frequency waves to establish a communication connection with the OBU device, and then read the unique identification information on the OBU device, such as the serial number in the chip.
[0086] Step S3: The license plate image data and the device number data are processed and stored by the edge computing unit, thereby obtaining the license plate device storage data;
[0087] Specifically, for example, the edge computing unit uses computer vision technology to process license plate image data, such as license plate detection, segmentation, and character recognition. At the same time, it associates the OBU device number data with the license plate image data and establishes a database to store this information. This database can be implemented using a traditional relational database management system (RDBMS) such as MySQL or a NoSQL database such as MongoDB.
[0088] Step S4: Obtain the number data of the vehicle-mounted equipment to be detected and the license plate image data to be detected through the intelligent roadside equipment, and use the license plate device storage data and the number data of the vehicle-mounted equipment to be detected to deeply restore the license plate image data to be detected through the edge computing unit, so as to obtain the license plate data of the masked vehicle.
[0089] Specifically, for example, deep learning models: use convolutional neural networks (CNNs) or more advanced architectures, such as convolutional recurrent neural networks (CRNNs) or Transformers, to perform feature extraction and character recognition on the license plate image data to be detected. For example, these models can be built and trained using PyTorch or TensorFlow to recognize characters on license plates. Image processing: Before the deep learning model, traditional computer vision techniques can be used to preprocess the image, including denoising, enhancement, and image correction. For example, the OpenCV library can be used for image processing. Data matching: by matching the number data of the on-board device to be detected with the data stored in the license plate device, to ensure that the device number matches the license plate correctly. License plate data integrity detection: Detect the integrity of the license plate image data to be detected, such as whether there is occlusion or missing, by comparing the license plate image with the stored license plate image, where the image similarity metric of the structural similarity index (SSIM) is used. Image mapping and deep restoration: If the license plate data is incomplete, image mapping techniques, such as pixel matching or image registration-based methods, can be used to deeply restore the license plate data, or calculate the geometric transformation matrix to restore the shape of the license plate.
[0090] In the present invention, by combining license plate image acquisition and on-board ETC system data acquisition, the method can improve the recognition accuracy of covered vehicles. Traditional license plate recognition systems are interfered by factors such as license plate stains, occlusions, and backlighting, while this method can improve the reliability of recognition with the assistance of ETC system data. Through the deep restoration technology in step S4, even if the license plate image quality is poor or there is occlusion, the license plate data can be restored. This helps to solve problems that are difficult to handle with visual analysis, such as covered or damaged license plates. Since the license plate image acquisition and equipment number acquisition in the method are carried out simultaneously, real-time or efficient recognition can be achieved without the need for additional parking waiting time. This helps to improve the efficiency of road traffic management. By associating the license plate image data with the on-board ETC system data, it can be ensured that the license plate information is associated with the equipment number of a specific vehicle.
[0091] Preferably, the license plate image data includes first license plate image data and second license plate image data, the license plate acquisition camera includes a first license plate acquisition camera and a second license plate acquisition camera, and step S1 specifically includes:
[0092] Controlling the first vehicle acquisition camera to acquire first vehicle image data at a first vehicle acquisition angle, thereby acquiring first vehicle image data;
[0093] Specifically, for example, the first vehicle acquisition camera is controlled to acquire the first vehicle image data at a first vehicle acquisition angle, such as 15° from bottom to top, such as a first exposure parameter and a first zoom parameter, so as to acquire the first vehicle image data;
[0094] The second vehicle acquisition camera is controlled to acquire the second vehicle image data at a second vehicle acquisition angle, thereby acquiring the second vehicle image data.
[0095] Specifically, for example, the second vehicle acquisition camera is controlled to acquire the second vehicle image data at a second vehicle acquisition angle, such as 15° from top to bottom, such as a second exposure parameter and a second zoom parameter, so as to obtain the second vehicle image data.
[0096] Specifically, for example, high-resolution industrial cameras, such as Basler ace cameras, are used, which can be controlled by a computer for precise image acquisition. These cameras are usually equipped with lenses suitable for different lighting conditions. The first car and the second car are in different lanes or positions. Therefore, the angle of the camera needs to be adjusted to ensure the correct acquisition of the license plate image of each car.
[0097] In the present invention, by using a first license plate acquisition camera and a second license plate acquisition camera, image data of two vehicles are acquired at the same time, or license plate data of vehicles are acquired from different angles, thereby reducing the error problem caused by illumination or angle, and improving the efficiency of data acquisition, especially in high-flow traffic scenes, license plate images can be acquired more quickly. Since the images of two vehicles are acquired at the same time, the influence of vehicle interference in the traffic flow on data acquisition is reduced. Even if multiple vehicles pass by at the same time, their license plate image data can be acquired effectively. For traffic monitoring and management, especially in scenes where multiple vehicles need to be identified in real time, acquiring image data of multiple vehicles at the same time can help improve efficiency.
[0098] Preferably, step S2 specifically includes:
[0099] Step S21: Communicate with the vehicle-mounted ETC system device through the intelligent roadside device installed on the roadside, thereby obtaining device communication data;
[0100] Specifically, for example, intelligent roadside equipment communicates with on-board ETC system equipment to obtain equipment communication data. Communication can be achieved through a series of protocols and technical means, such as on-board communication standards (such as DSRC, C-V2X), network protocols (such as TCP / IP). Numerical support includes specific values of performance parameters such as communication success rate and communication delay.
[0101] Step S22: Establish a connection between the intelligent roadside device and the vehicle-mounted ETC system device according to the device communication data, and collect the vehicle-mounted ETC system device number through the intelligent roadside device to obtain the primary device number data;
[0102] Specifically, for example, based on the device communication data, a connection is established between the intelligent roadside device and the on-board ETC system device, which is achieved through secure communication technologies such as TLS / SSL security protocols and identity authentication mechanisms. The collection of primary device number data includes sending a request to the OBU device through the RSU, and then the OBU device responds with its unique device number, such as communication success rate and response time.
[0103] Step S23: performing error detection and correction on the primary device number data, thereby obtaining revised device number data;
[0104] Specifically, for example, error detection and correction is performed on the primary device number data obtained from the OBU device. This can be achieved using error correction codes (such as Hamming codes or RS codes) to ensure the integrity and accuracy of the data. The numerical support includes performance parameters of the error correction code, such as error correction capability and error detection rate.
[0105] Step S24: storing the corrected device number data, thereby obtaining the device number data.
[0106] Specifically, for example, the corrected OBU data is stored for subsequent use, which is achieved using a database, a file system or other data storage technology, wherein the parameter settings include the capacity and storage speed of the data storage.
[0107] The present invention utilizes the communication data of the on-board ETC system, establishes a connection with the communication of the ETC system, can obtain the OBU device number associated with a specific vehicle, associates the license plate information with the OBU device number of the vehicle, and ensures the consistency and accuracy of the data. In step S23, the primary device number data is error detected and corrected, which helps to ensure the accuracy of the device number, can improve the accurate identification of the vehicle, and reduce the misidentification caused by the device number error. Step S24 stores the corrected OBU data to form the license plate device storage data, so that the system can quickly access and retrieve the OBU data associated with the vehicle, thereby improving efficiency in the license plate information recognition process. By establishing a connection with the OBU device and correcting the device number data, the reliability of vehicle identification can be improved, which is very important for ensuring that the correct license plate information is associated with a specific vehicle.
[0108] Preferably, in step S21, the communication is encrypted by using an intelligent roadside device authentication encryption value calculation formula, wherein the intelligent roadside device authentication encryption value calculation formula is specifically:
[0109] ;
[0110] Authenticate encrypted values for intelligent roadside equipment, is the communication data quantity data, is the communication data index item, For the Intelligent roadside equipment communication data, is the communication time range data, is the communication data change rate, is the communication time data, is the communication propagation angle data, To communicate relative angle data, Load data for intelligent roadside equipment, Encode data for intelligent roadside equipment data.
[0111] The present invention constructs a calculation formula for the authentication encryption value of intelligent roadside equipment, which provides stable and reliable data support in intelligent Internet of Things, especially in the encryption communication calculation of intelligent roadside equipment. It is the encrypted value for intelligent roadside equipment authentication, representing the final encrypted value for authentication. For the The communication data of intelligent roadside equipment represents a part of the communication data. Communication propagation angle data, indicating the direction of the communication device or the propagation direction of the signal. For example, if a device needs to adjust its antenna direction to communicate with another device, Indicates the direction angle of the antenna. The value is usually expressed in degrees or radians, depending on the specific standard used. Communication relative angle data, used to indicate the direction or angle related to communication. The specific meaning of also depends on the specific application scenario. In communication, Indicates the incident angle of the communication signal, or the relative direction between devices. For example, if two vehicles need to communicate with each other, Indicates the relative direction of one vehicle to another. The value of is usually expressed in degrees or radians. The present invention can improve the security of communications. This helps reduce the risk of unauthorized access and data leakage. The calculation formula includes consideration of the rate of change of the communication data, which can be used to detect inconsistencies or tampering attempts of the data. If the data changes or is tampered with, the authentication will fail, thereby protecting the integrity of the data. The parameters in the formula and The load and data encoding characteristics of the communication device are taken into account, and the performance and characteristics of the device are taken into account during the authentication process, so as to better adapt to different types of devices. The invention aims to improve the security and reliability of intelligent roadside device authentication, while taking into account the characteristics of the communication device, thereby enhancing the performance and security of the entire communication system. This is essential to ensure secure communication between intelligent roadside devices and the integrity of data, especially in scenarios involving vehicle communications and intelligent transportation systems.
[0112] Preferably, step S23 is specifically:
[0113] Step S231: creating an original device number according to the primary device number data, thereby obtaining the original device number data;
[0114] Specifically, for example, an encryption algorithm or a hash function, such as SHA-256, is used to convert the primary device number into the original device number. For example, for the primary device number 12345, the original device number "6f82b8a15b8a8e2350c64e39653aa589" may be generated.
[0115] Step S232: creating an error detection area and performing data replication according to the original device number data, thereby obtaining the error detection device number data;
[0116] Specifically, for example, error detection area definition: using defined rules and algorithms, the area containing errors in the original device number is delineated. For example, the device number is divided into different blocks or areas, some of which are prone to errors. Data replication: copying the original device number data for subsequent error correction processing is achieved by creating a copy or backup data.
[0117] More importantly, specifically, for example, detection environment data is obtained, and interference estimation is performed based on the detection environment data, thereby obtaining environmental interference estimation data, wherein the environmental interference estimation data includes composite environment interference estimation data, single environment interference estimation data and safe environment estimation data; when it is determined that the environmental interference estimation data is composite environment interference estimation data, the original device number data is divided into composite areas, thereby obtaining OBU error check area data; when it is determined that the environmental interference estimation data is single environment interference estimation data, error check area data is generated based on the single environment interference estimation data, and the original device number data is divided into a single area based on the error check area data, thereby obtaining OBU error check area data; when it is determined that the environmental interference estimation data is safe environment estimation data, the original device number data is divided into error check areas based on preset error check area data generation parameters, thereby obtaining OBU error check area data; data replication is performed on the original device number data based on the OBU error check area data, thereby obtaining error detection device number data.
[0118] Specifically, for example, when it is determined that the environmental interference prediction data is composite environmental interference prediction data, the original equipment number data is divided into composite areas to obtain OBU error check area data, wherein the composite area division is adaptively adjusted and generated according to the parameters of the interference factors, and voting is performed in the subsequent error correction process to generate error correction results.
[0119] Specifically, for example, environmental interference prediction data acquisition: in actual scenarios, environmental data is obtained through various sensors (such as weather sensors, light sensors, cameras or vehicle / road-mounted signal sensors), for example: composite environmental interference prediction data: consider multiple factors, such as weather, lighting, and traffic conditions, and comprehensively evaluate the impact on recognition. Single environmental interference prediction data: perform interference prediction for a single factor, such as obstacles, weather, or insufficient signal strength. Safe environment prediction data: focus on interference in traffic and communication, such as road conditions and intersections with relatively suitable communication environments. Processing based on interference prediction data: adopt corresponding processing strategies based on different types of environmental interference prediction data: If the environmental interference prediction data is composite environmental interference prediction data: perform composite area division, and divide the device number data into different areas for error checking. If the environmental interference prediction data is single environmental interference prediction data: generate error check area data based on the single environmental interference prediction data, and then divide the device number data into corresponding areas. If the environmental interference estimation data is the safety environment estimation data: Generate parameters according to the preset error check area data, and then divide the device number data into corresponding areas. Error detection: Detect and correct the original device number data according to the error check area data to obtain the error detection device number data.
[0120] Step S233: segmenting the error detection device number data to obtain device number segmentation data;
[0121] Specifically, for example, the segmentation algorithm: The error detection device number data is divided into different characters or blocks using a segmentation algorithm. This can be achieved by identifying gaps between characters or using a segmentation symbol.
[0122] Step S234: extracting character features and pixel density features from the device number segmentation data, thereby obtaining device number character feature data and device number pixel density feature data;
[0123] Specifically, for example, character feature extraction: use image processing technology, such as edge detection algorithm (such as Sobel operator), to extract the shape, outline and other features of the character. Pixel density feature extraction: calculate the pixel density of each character, that is, the proportion of pixels occupied by the character, by counting the pixels of the character image.
[0124] Step S235: performing a first recognition calculation on the device number character feature data and the device number pixel density feature data, thereby obtaining device number error detection position data;
[0125] Specifically, for example, error detection algorithm: using machine learning algorithms or rule engines, such as convolutional neural networks (CNN) or conditional judgments, to determine which characters contain errors. Error position marking: marking the position of the detected erroneous characters for subsequent correction processing.
[0126] Step S236: performing shape edge curvature feature extraction and projection density feature extraction on the device number segmentation data, thereby obtaining shape edge curvature feature data and projection density feature data;
[0127] Specifically, for example, shape edge curvature extraction: Use image processing techniques such as the Canny edge detection algorithm to detect the edge of the device number. Then, by calculating the curvature or shape features of the edge, the shape edge curvature feature data is extracted. Projection density feature extraction: By projecting the device number image in the horizontal and vertical directions, the distribution of pixel density can be calculated. This can be used to extract projection density feature data. Use numerical methods to calculate the curvature of each pixel. Curvature indicates the degree of curvature of a curve at a specific point. For edges in an image, the curvature can be calculated using the following formula: Curvature (K) = (d²y / dx²) / (1+(dy / dx)²)^(3 / 2), where d²y / dx² represents the second-order derivative and dy / dx represents the slope. This formula is used to calculate the curvature value at each pixel. Use edge detection algorithms (such as Sobel, Canny) to detect edges in the device number segmentation data. This will help determine which pixels contain edge information. For detected edge pixels, calculate their curvature values. Then, statistical characteristics of these curvature values can be calculated, such as the mean curvature, maximum curvature, and minimum curvature.
[0128] Step S237: performing a second identification calculation on the shape edge curvature feature data and the projection density feature data, thereby obtaining device number error detection content data;
[0129] Specifically, for example, machine learning or deep learning models: Based on the extracted feature data, a second recognition calculation is performed using machine learning or deep learning models, including neural network models, logistic regression, and support vector machines. Feature fusion: Combining shape edge curvature features and projection density features to obtain more accurate recognition results, which is achieved through feature fusion technology.
[0130] Step S238: performing correction processing on the device number segmentation data according to the device number error detection position data and the device number error detection content data, thereby obtaining corrected device number data;
[0131] Specifically, for example, the error detection position data generally includes information indicating which characters or bits are erroneous. For example, it can be a binary sequence, in which each bit indicates whether the corresponding position is erroneous, 1 indicates an error, and 0 indicates correctness. Such as [0,1, 0, 0, 1]. The error detection content data includes information indicating the corrected character value or bit value. For example, it can be a character sequence, in which each character represents the corrected value of the corresponding position. Such as ['A', 'B', 'C', 'D', 'E']. According to the error detection position data and the error detection content data, the erroneous characters in the device number segmentation data are replaced. According to the above example, the characters in the corresponding positions are replaced as follows: original OBU data: ABXDE, corrected OBU data: ABCDE.
[0132] Step S239: Perform character combination on the corrected device number data to obtain the corrected device number data.
[0133] Specifically, for example, character combination: In this step, the corrected characters are combined into a complete device number in the order in which they appear in the device number. This is a simple operation, just connect the characters together in the correct order. Corrected OBU data: ABCDE Corrected device number: ABCDE.
[0134] In the present invention, step S231 converts the device number data from the primary state to the original state by creating the original device number, which is used for comparison with subsequent error detection and correction. Step S232 creates an error detection area for determining the part containing the error and copies the data for later use, which helps to narrow the scope of the error and reduce the computational complexity of error detection. Step S233 divides the device number into smaller parts so that each part can be individually error detected and corrected. This improves the accuracy and reliability of the system. Step S234 extracts character features and pixel density features from the device number segmentation data, providing information about the shape and density of the character, which helps with subsequent error detection. Step S235 uses the character features and pixel density features to perform a first recognition calculation to determine the location of the error detection, which helps to determine the location where the error exists in the device number. Step S236 extracts the shape edge curvature features and projection density features, which can provide more information about the character shape for further error detection. Step S237 uses the shape edge curvature features and projection density features to perform a second recognition calculation to determine the content of the error detection, which helps to determine the specific character or content of the error. Step S238 corrects the device number segmentation data according to the position and content of the error detection, and combines the corrected partial characters into the revised OBU data, thereby ensuring the accuracy of the device number.
[0135] Preferably, step S3 specifically includes:
[0136] Step S31: The license plate image data and the device number data are transmitted to the edge computing unit, and the edge computing unit performs parsing to obtain the license plate image parsing data and the device number parsing data;
[0137] Specifically, for example, data transmission protocol: Use communication protocols such as HTTP and TCP / IP to transmit data from the collection point to the edge computing unit. Data parsing algorithm: Use parsing algorithms such as image parsing algorithms and device number parsing algorithms in the edge computing unit to parse the received data to extract useful information.
[0138] Step S32: performing intelligent image correction processing on the license plate image analysis data using preset license plate image standard parameters, thereby obtaining license plate image correction data;
[0139] Specifically, for example, image correction algorithm: Use image processing techniques, such as perspective transformation or affine transformation, to correct the image according to the characteristics of the license plate image so that it meets the preset license plate image standard parameters. License plate image standard parameters: Preset a set of license plate image standard parameters, including image size, resolution, and rotation angle, to ensure that the corrected image meets specific standards.
[0140] Step S33: Associating and merging the license plate image correction data and the device number parsing data, thereby obtaining license plate device number associated merged data;
[0141] Specifically, for example, data association rules: using association rules, such as matching the timestamp of the license plate image with the timestamp of the device number, the corresponding image and device number are associated. Data merging algorithm: merging the associated license plate image and device number data to create license plate device number associated merged data.
[0142] Step S34: the license plate device number associated merged data is access-controlled marked and stored through a preset multi-level data access control strategy, thereby obtaining the license plate device storage data.
[0143] Specifically, for example, data access control policy: Preset multi-level data access control policy to determine who has the right to access the license plate device number associated merged data. This can include role-based access control and access time restrictions. Data annotation: Annotate the data to ensure the integrity and security of the data, including adding timestamps and digital signatures. Data storage: Store the license plate device number associated merged data. You can choose to use a database, cloud storage or local storage for subsequent query and analysis. Such as access control annotation: According to the multi-level data access control policy, the edge computing unit will add corresponding tags to each data record, indicating its level and allowed operations. For example: License plate image data is marked as level 2 and is only accessible to vehicle owners and law enforcement agencies. Location information data is marked as level 1 and is only accessible to vehicle owners. Vehicle identification data is marked as level 2 and is accessible to vehicle owners and road management departments.
[0144] In the present invention, step S31 is responsible for transmitting the license plate image data and the equipment number data to the edge computing unit, and parsing the data, which helps to associate the two and ensure that the subsequent steps can correctly process the data. Step S32 performs intelligent image correction processing on the license plate image parsing data to correct the license plate image to a standardized format according to the preset license plate image standard parameters, which helps to improve the accuracy and readability of subsequent license plate information. Step S33 associates and merges the license plate image correction data with the equipment number parsing data to ensure that the license plate information is correctly matched with the vehicle's equipment number, providing accurate data for subsequent access control and storage. Step S34 uses a multi-level data access control strategy to perform access control annotation and storage on the license plate equipment number associated merge data, ensuring that only authorized users can access and use the data, thereby improving data security and privacy protection.
[0145] Preferably, the intelligent image correction processing in step S32 is performed by using an intelligent image correction processing calculation formula, wherein the intelligent image correction processing calculation formula is specifically:
[0146] ;
[0147] Correction data for license plate images, is the lateral blur corrected for the license plate image, is the longitudinal blur correction of the license plate image, is the horizontal coordinate point of the license plate image, is the horizontal coordinate point of the center point of the license plate image, is the longitudinal coordinate point of the license plate image, is the longitudinal coordinate point of the center point of the license plate image, To correct the fluctuation period term, is the distance correction factor, is the original license plate image intensity distribution data.
[0148] The present invention constructs an intelligent image correction processing calculation formula, which corrects the license plate image to improve its clarity and readability. is the horizontal coordinate point of the license plate image and It is the longitudinal coordinate point of the license plate image, indicating the coordinate point on the image, and is used to specify the corrected pixel position. The lateral blur corrected for the license plate image and The longitudinal blurriness corrected for the license plate image represents the standard deviation of the image, which is used to control the lateral and longitudinal blurriness of the image correction. is the horizontal coordinate point of the center point of the license plate image and The longitudinal coordinate point of the center point of the license plate image represents the coordinates of the center point of the image and is used to specify the center position of the correction. The correction fluctuation period term represents the wavelength of the light wave and is used to control the fluctuation period in the correction. It is the correction distance factor, which represents the distance on the image and is used to specify the correction distance factor. is the intensity distribution data of the original license plate image, which represents the intensity distribution of the original license plate image. This formula describes the process of intelligent image correction. To control the blur of the image, the cosine function To introduce the wave effect, as well as the image intensity distribution to represent the characteristics of the original license plate image. The correction term in the formula can correct the distortion and blur in the license plate image and improve the clarity and quality of the license plate image. This helps to recognize the license plate information more accurately. By adjusting parameters such as the horizontal and vertical blur ( and )、Fluctuation period term( ) and the correction distance factor ( ), the method can adapt to different lighting conditions, angles and distances, thereby improving the license plate recognition capability in various environments. By using the cosine function to introduce the wave effect, the misrecognition caused by the reflection, distortion or occlusion of the license plate image can be reduced, and the stability of the license plate recognition can be improved. The present invention improves the accuracy and stability of license plate information recognition.
[0149] Preferably, step S4 is specifically:
[0150] Step S41: Acquire the number data of the vehicle-mounted device to be detected and the image data of the license plate to be detected through the intelligent roadside device;
[0151] Specifically, for example, device number data collection: RSU establishes a connection with the vehicle-mounted ETC system device and uses a communication protocol (such as HTTP, TCP / IP) to obtain the vehicle-mounted device number data to be detected. License plate image data collection: RSU collects images through the license plate recognition camera and then transmits the license plate image data to be detected to the edge computing unit.
[0152] Step S42: verifying and correcting the vehicle-mounted device number data through the edge computing unit, thereby obtaining error correction data of the vehicle-mounted device number to be detected;
[0153] Specifically, for example, error detection algorithms: Error detection and correction algorithms, such as cyclic redundancy check (CRC) or Hamming code, are used to detect and correct errors in the device number data. For example, cyclic redundancy check (CRC): Working principle: CRC is a polynomial-based check method that detects errors by treating the data as a binary polynomial and calculating polynomial division. The sender calculates the CRC check value of the data and appends it to the data for transmission. The receiver also calculates the CRC check value of the received data and then compares it with the CRC value provided by the sender. If the two values do not match, it indicates that there is an error in the data. Application: In the processing of device number data, CRC can be used to detect any unexpected changes in data transmission. The OBU device calculates the CRC check value before sending the data and appends it to the data. The receiver also calculates the CRC check value of the received data and then compares it with the CRC value provided by the sender. If it does not match, the receiver can request retransmission or take other measures to correct the error.
[0154] Step S43: matching the data stored in the license plate device with the error correction data of the vehicle-mounted device number to obtain matching data of the license plate to be detected;
[0155] Specifically, for example, database query: using a database management system such as MySQL or MongoDB, a query operation is performed to find a matching device number in the license plate device storage data.
[0156] Step S44: performing a license plate data integrity test on the license plate image data to be detected, thereby obtaining license plate data integrity test data, thereby obtaining license plate data integrity test data, wherein the license plate data integrity test data includes complete license plate data and missing license plate data;
[0157] Specifically, for example, image processing algorithms: Use image processing algorithms to analyze license plate images to detect whether there is a missing or otherwise damaged license plate. Data integrity indicators: Define data integrity indicators, such as the number of license plate characters and character positions, to evaluate the integrity of the license plate image.
[0158] Step S45: if the license plate data integrity detection data is determined to be complete license plate data, the license plate image data to be detected is determined to be the license plate data of the masked vehicle;
[0159] Specifically, for example, if the license plate data integrity detection data indicates that the license plate is intact, then the license plate image data to be detected can be determined as the license plate data of a masked vehicle.
[0160] Step S46: determining that the license plate data integrity detection data is license plate missing data, then performing image mapping on the license plate image data to be detected according to the license plate image data in the license plate device storage data, thereby obtaining image mapping data;
[0161] Specifically, for example, an image mapping algorithm: Use an image mapping algorithm, such as perspective transformation, to map the license plate image to be detected onto a standard license plate template, thereby restoring the shape and content of the license plate.
[0162] Step S47: performing confidence calculation on the image mapping data, thereby obtaining image mapping confidence data;
[0163] Specifically, for example, the confidence calculation is performed on the image mapping data to determine whether the license plate data of the masked vehicle can be accurately identified. Based on the features and the matching results, a confidence score is calculated to measure the credibility of the license plate information.
[0164] Specifically, for example, vehicle identification feature data (VIF): used to identify the characteristics of vehicle type and brand. License plate image quality feature data (QIF): used to evaluate the clarity and quality of license plate images. Masked license plate feature data (MF): used to detect whether there are signs of masked license plates, such as occlusion and reflection. A comprehensive confidence score (CS) is defined as follows: Among them: w1, w2, w3 are weights used to adjust the contribution of different features to confidence, which are the first confidence weight coefficient, the second confidence weight coefficient, and the third confidence weight coefficient respectively. VIF is the score of vehicle identification feature data, which is calculated using machine learning or image processing models. QIF is the score of license plate image quality feature data, which is calculated based on factors such as image clarity and contrast. MF is the score of masked feature data, which is calculated by an algorithm that detects signs of masking.
[0165] Step S48: when it is determined that the image mapping confidence data is the first image mapping confidence data, the license plate image data in the license plate device storage data is determined as the masked vehicle license plate data;
[0166] Specifically, for example, once the confidence calculation is completed, the system will make a decision based on the result. If the confidence of the image mapping is high, indicating that the license plate information is accurate, then the license plate image data in the license plate device storage data can be determined as the license plate data of the masked vehicle. This means that the system believes that the mapped image provides accurate license plate information.
[0167] Step S49: When it is determined that the image mapping confidence data is the second image mapping confidence data, the license plate image data to be detected is deeply restored to obtain the license plate data of the masked vehicle.
[0168] Specifically, for example, once the confidence calculation is completed, the system will make a decision based on the result. If the confidence of the image mapping is high, indicating that the license plate information is accurate, then the license plate image data in the license plate device storage data can be determined as the license plate data of the masked vehicle. This means that the system believes that the mapped image provides accurate license plate information.
[0169] Specifically, for example, the license plate image data to be detected is recognized and restored by a preset license plate image restoration recognition model, thereby obtaining the license plate data of the masked vehicle. The license plate image restoration recognition model is generated by training the preset license plate image training data through a neural network model, a support decision tree algorithm, and an integrated learning algorithm.
[0170] In the present invention, step S41 is responsible for obtaining the vehicle-mounted device number data to be detected and the license plate image data to be detected. Step S42 ensures the accuracy of the device number by checking and correcting the vehicle-mounted device number data to be detected, thereby improving the reliability of vehicle identification. Step S43 matches the vehicle-mounted device number correction data with the license plate device storage data according to the vehicle-mounted device number correction data to be detected, thereby ensuring the correct correspondence between the device number and the license plate information. Step S44 performs integrity detection on the license plate image data to be detected to determine whether the license plate is complete and distinguish between masked vehicles and other situations, such as partial occlusion or damage of the license plate. Step S46 performs image mapping on the license plate image data to be detected, and step S47 calculates the confidence of the image mapping, determines the credibility of the image mapping, and then determines whether deep restoration is needed. Step S49 performs deep restoration on the license plate image data to be detected to obtain the license plate data of the masked vehicle, which plays a key role when the license plate data is missing or incomplete. The present invention performs multi-level identification and verification on the license plate information of the masked vehicle, thereby improving the accuracy and credibility of vehicle identification.
[0171] Preferably, the deep restoration in step S49 is performed through an image restoration process using a vehicle license plate deep restoration image model, wherein the steps of constructing the vehicle license plate deep restoration image model are specifically as follows:
[0172] Step S491: Acquire standard covered vehicle license plate image data and corresponding standard covered vehicle license plate data;
[0173] Specifically, for example, a series of standard masked vehicle license plate images and corresponding license plate data are selected from an existing license plate database, and these license plate images include license plates of different vehicle models, angles, and lighting conditions.
[0174] Step S492: extracting features from the standard covered-plate vehicle license plate image data, thereby obtaining the standard covered-plate vehicle license plate image feature data;
[0175] Specifically, for example, feature extraction includes but is not limited to: linear operators, Gaussian operators, and pre-trained deep learning models for feature extraction.
[0176] Step S493: performing complexity processing on the image feature data of the standard covered vehicle license plate, thereby obtaining image feature complexity data;
[0177] Specifically, for example, to calculate the complexity of the extracted image features, techniques such as feature histogram and spatial pyramid pooling can be used to quantify the complexity of the image features.
[0178] Step S494: generating image feature convolution layer data according to the image feature complexity data, and performing convolution calculation on the standard masked vehicle license plate image feature data using the image feature convolution layer data, thereby obtaining the standard masked vehicle license plate image convolution layer data;
[0179] Specifically, for example, the convolutional layer structure of the convolutional neural network is constructed according to the image feature complexity data, including the number and size of convolution kernels. These convolution kernels are used to perform convolution calculations on the standard masked vehicle license plate image feature data to generate convolutional layer data.
[0180] Specifically, for example, input: standard masked vehicle license plate image feature data (e.g., 224x224 pixel feature map) and generated feature complexity data. The feature map is convolved using a convolution kernel, but the weights of the convolution kernel are affected by the feature complexity data. Areas with high feature complexity will affect the value of the weight, thereby changing the result of the convolution calculation. Output: standard masked vehicle license plate image convolution layer data, where each pixel value is adjusted by the feature complexity data.
[0181] Specifically, for example, the convolution kernel weights: for areas with high feature complexity, the weight values are lower, and for areas with low feature complexity, the weight values are higher. Output feature map size: Same as the input feature map, but the pixel values have been adjusted according to the convolution kernel and feature complexity.
[0182] Step S495: performing pooling layer calculation on the convolution layer data of the standard masked vehicle license plate image, thereby obtaining the pooling layer data of the standard masked vehicle license plate image;
[0183] Specifically, for example, a pooling operation, such as maximum pooling or average pooling, is performed on the convolutional layer data to reduce the data dimension and extract the most important features.
[0184] Step S496: performing full connection calculation on the pooling layer data of the standard masked vehicle license plate image, thereby obtaining full connection weight data of the standard masked vehicle license plate image;
[0185] Specifically, for example, a fully connected neural network layer is constructed, and the pooling layer data of the standard masked vehicle license plate image is used to perform fully connected calculations to generate fully connected weight data.
[0186] Step S497: Annotate the fully connected weighted data of the standard masked vehicle license plate image according to the standard masked vehicle license plate data, thereby obtaining a vehicle license plate deep restoration image model.
[0187] Specifically, for example, the fully connected weighted data is annotated using standard masked vehicle license plate data. This may include training a classifier or regression model to map the standard masked vehicle license plate image data to the corresponding license plate data. After the model is obtained, it can be used to perform deep restoration of the license plate image data to be detected.
[0188] The model of the present invention can be used to restore the license plate image of the covered vehicle, and the license plate information can be effectively restored even in the case of blur, noise or damage. This helps to improve the accuracy of vehicle recognition, especially in security monitoring and traffic management. The model uses deep learning technology, especially convolutional neural network (CNN) and fully connected network, to extract features and perform complex calculations on the image, and can learn higher-level image features, thereby improving the restoration effect. The license plate image is affected by different lighting, weather conditions and angles, so higher-level features are required to identify the license plate in different situations. The introduction of image feature convolution layer data can improve the model's adaptability to these complex scenes and reduce the probability of recognition errors. The convolution calculation in this step can effectively fuse the image features with the standard covered vehicle license plate image feature data. This fusion helps to improve the generalization performance of the deep learning model, so that it can process license plate images of different types and styles, not just for specific situations. By using standard covered vehicle license plate image data and corresponding standard covered vehicle license plate data, the model can be effectively trained and labeled. This ensures that the model can accurately restore the license plate image during the restoration process, especially in terms of license plate image diversity.
[0189] Preferably, a device for recognizing license plate information of a covered vehicle based on visual technology is used to perform the method for recognizing license plate information of a covered vehicle based on visual technology as described above, and the device for recognizing license plate information of a covered vehicle based on visual technology comprises:
[0190] The license plate image acquisition module is used to acquire the license plate image through a preset license plate acquisition camera, thereby obtaining the license plate image data;
[0191] The equipment number data collection module is used to collect the equipment number of the vehicle-mounted ETC system through the intelligent roadside equipment installed on the roadside, so as to obtain the equipment number data;
[0192] A license plate device number storage module is used to process and store the license plate image data and device number data through the edge computing unit, thereby obtaining the license plate device storage data;
[0193] The retrieval and deep restoration module is used to obtain the numbering data of the on-board equipment to be detected through the intelligent roadside equipment, and use the data stored in the license plate device to retrieve and deeply restore the numbering data of the on-board equipment to be detected through the edge computing unit, so as to obtain the license plate data of the masked vehicle.
[0194] The beneficial effect of the present invention is that the method can comprehensively collect license plate image data and vehicle ETC system equipment number data, thereby ensuring the multi-dimensional information collection of the vehicle and providing sufficient data support for subsequent identification and verification. The present invention includes the collection and verification process of the equipment number, ensuring the accuracy of the equipment number, ensuring that the vehicle and its corresponding OBU equipment are correctly matched, and avoiding misidentification. The license plate image data and equipment number data are transmitted to the edge computing unit and stored as license plate device storage data. The multi-level data storage adopted allows the data to be effectively managed for subsequent query and identification. The deep restoration technology in the present invention is applied to the license plate image data of the vehicle to be detected, and the license plate information can still be restored when the license plate image is blocked, defaced, or backlit. Through the deep restoration technology, the method can highly reliably identify the license plate information of the masked vehicle, which is of great significance in improving traffic management, safety monitoring, and crime prevention and control, and helps to reduce illegal acts and improve the level of public security. The edge computing unit allows high efficiency in real time and can process more vehicles and data, so as to cope with peak traffic flow and the growing number of vehicles.
[0195] Therefore, from any point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the attached application documents rather than the above description, and it is intended that all changes falling within the meaning and scope of equivalent elements of the application documents are included in the present invention.
[0196] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.
Claims
1. A method for recognizing license plate information of covered vehicles based on visual technology, characterized in that: The following steps are involved: Step S1: Capturing the license plate image through a preset license plate acquisition camera, thereby obtaining license plate image data; Step S2 comprises: Step S21: Communicate with the vehicle-mounted ETC system device through the intelligent roadside device installed on the roadside, thereby obtaining device communication data; Step S22: Establish a connection between the intelligent roadside device and the vehicle-mounted ETC system device according to the device communication data, and collect the vehicle-mounted ETC system device number through the intelligent roadside device to obtain the primary device number data; Step S23 includes: Step S231: creating an original device number according to the primary device number data, thereby obtaining the original device number data; Step S232: creating an error detection area and performing data replication according to the original device number data, thereby obtaining the error detection device number data; Step S233: segmenting the error detection device number data to obtain device number segmentation data; Step S234: extracting character features and pixel density features from the device number segmentation data, thereby obtaining device number character feature data and device number pixel density feature data; Step S235: performing a first recognition calculation on the device number character feature data and the device number pixel density feature data, thereby obtaining device number error detection position data; Step S236: performing shape edge curvature feature extraction and projection density feature extraction on the device number segmentation data, thereby obtaining shape edge curvature feature data and projection density feature data; Step S237: performing a second identification calculation on the shape edge curvature feature data and the projection density feature data, thereby obtaining device number error detection content data; Step S238: performing correction processing on the device number segmentation data according to the device number error detection position data and the device number error detection content data, thereby obtaining corrected device number data; Step S239: performing character combination on the corrected device number data, thereby obtaining the corrected device number data; Step S24: storing the corrected device number data, thereby obtaining the device number data; Step S3: The license plate image data and the device number data are processed and stored by the edge computing unit, thereby obtaining the license plate device storage data; Step S4: Obtain the number data of the vehicle-mounted equipment to be detected and the license plate image data to be detected through the intelligent roadside equipment, and use the license plate device storage data and the number data of the vehicle-mounted equipment to be detected to deeply restore the license plate image data to be detected through the edge computing unit, so as to obtain the license plate data of the masked vehicle.
2. The method according to claim 1, characterized in that The license plate image data includes first license plate image data and second license plate image data, the license plate acquisition camera includes a first license plate acquisition camera and a second license plate acquisition camera, and step S1 is specifically: Controlling the first vehicle acquisition camera to acquire first vehicle image data at a first vehicle acquisition angle, thereby acquiring first vehicle image data; The second vehicle acquisition camera is controlled to acquire the second vehicle image data at a second vehicle acquisition angle, thereby acquiring the second vehicle image data.
3. The method according to claim 1, characterized in that In step S21, the communication is encrypted by the intelligent roadside device authentication encryption value calculation formula, wherein the intelligent roadside device authentication encryption value calculation formula is specifically: ; Authenticate encrypted values for intelligent roadside equipment, is the communication data quantity data, is the communication data index item, For the Intelligent roadside equipment communication data, is the communication time range data, is the communication data change rate, is the communication time data, is the communication propagation angle data, To communicate relative angle data, Load data for intelligent roadside equipment, Encode data for intelligent roadside equipment data.
4. The method according to claim 1, characterized in that: Step S3 is specifically as follows: Step S31: The license plate image data and the device number data are transmitted to the edge computing unit, and the edge computing unit performs parsing to obtain the license plate image parsing data and the device number parsing data; Step S32: performing intelligent image correction processing on the license plate image analysis data using preset license plate image standard parameters, thereby obtaining license plate image correction data; Step S33: Associating and merging the license plate image correction data and the device number parsing data, thereby obtaining license plate device number associated merged data; Step S34: the license plate device number associated merged data is access-controlled marked and stored through a preset multi-level data access control strategy, thereby obtaining the license plate device storage data.
5. The method according to claim 4, characterized in that In step S32, the intelligent image correction processing is performed by using an intelligent image correction processing calculation formula, wherein the intelligent image correction processing calculation formula is specifically: ; Correction data for license plate images, is the lateral blur corrected for the license plate image, is the longitudinal blur correction of the license plate image, is the horizontal coordinate point of the license plate image, is the horizontal coordinate point of the center point of the license plate image, is the longitudinal coordinate point of the license plate image, is the longitudinal coordinate point of the center point of the license plate image, To correct the fluctuation period term, is the distance correction factor, is the original license plate image intensity distribution data.
6. The method according to claim 1, characterized in that Step S4 is specifically as follows: Step S41: Acquire the number data of the vehicle-mounted device to be detected and the image data of the license plate to be detected through the intelligent roadside device; Step S42: verifying and correcting the vehicle-mounted device number data through the edge computing unit, thereby obtaining error correction data of the vehicle-mounted device number to be detected; Step S43: matching the data stored in the license plate device with the error correction data of the vehicle-mounted device number to obtain matching data of the license plate to be detected; Step S44: performing a license plate data integrity test on the license plate image data to be detected, thereby obtaining license plate data integrity test data, thereby obtaining license plate data integrity test data, wherein the license plate data integrity test data includes complete license plate data and missing license plate data; Step S45: if the license plate data integrity detection data is determined to be complete license plate data, the license plate image data to be detected is determined to be the license plate data of the masked vehicle; Step S46: determining that the license plate data integrity detection data is license plate missing data, then performing image mapping on the license plate image data to be detected according to the license plate image data in the license plate device storage data, thereby obtaining image mapping data; Step S47: performing confidence calculation on the image mapping data, thereby obtaining image mapping confidence data; Step S48: when it is determined that the image mapping confidence data is the first image mapping confidence data, the license plate image data in the license plate device storage data is determined as the masked vehicle license plate data; Step S49: When it is determined that the image mapping confidence data is the second image mapping confidence data, the license plate image data to be detected is deeply restored to obtain the license plate data of the masked vehicle.
7. The method according to claim 6, characterized in that In step S49, the deep restoration is performed through the vehicle license plate deep restoration image model to perform image restoration processing, wherein the steps for constructing the vehicle license plate deep restoration image model are specifically as follows: Step S491: Acquire standard covered vehicle license plate image data and corresponding standard covered vehicle license plate data; Step S492: extracting features from the standard covered-plate vehicle license plate image data, thereby obtaining the standard covered-plate vehicle license plate image feature data; Step S493: performing complexity processing on the image feature data of the standard covered vehicle license plate, thereby obtaining image feature complexity data; Step S494: generating image feature convolution layer data according to the image feature complexity data, and performing convolution calculation on the standard masked vehicle license plate image feature data using the image feature convolution layer data, thereby obtaining the standard masked vehicle license plate image convolution layer data; Step S495: performing pooling layer calculation on the convolution layer data of the standard masked vehicle license plate image, thereby obtaining the pooling layer data of the standard masked vehicle license plate image; Step S496: performing full connection calculation on the pooling layer data of the standard masked vehicle license plate image, thereby obtaining full connection weight data of the standard masked vehicle license plate image; Step S497: Annotate the fully connected weighted data of the standard masked vehicle license plate image according to the standard masked vehicle license plate data, thereby obtaining a vehicle license plate deep restoration image model.
8. A device for recognizing license plate information of covered vehicles based on visual technology, characterized in that: Used to perform the method for recognizing license plate information of covered vehicles based on vision technology as claimed in claim 1, the device for recognizing license plate information of covered vehicles based on vision technology comprises: The license plate image acquisition module is used to acquire the license plate image through a preset license plate acquisition camera, thereby obtaining the license plate image data; The equipment number data collection module is used to collect the equipment number of the vehicle-mounted ETC system through the intelligent roadside equipment installed on the roadside, so as to obtain the equipment number data; A license plate device number storage module is used to process and store the license plate image data and device number data through the edge computing unit, thereby obtaining the license plate device storage data; The retrieval and deep restoration module is used to obtain the numbering data of the on-board equipment to be detected through the intelligent roadside equipment, and use the data stored in the license plate device to retrieve and deeply restore the numbering data of the on-board equipment to be detected through the edge computing unit, so as to obtain the license plate data of the masked vehicle.
