License plate recognition method and device

Abstract

The application provides a license plate recognition method and device, the license plate recognition method comprising: acquiring a first vehicle image; performing first convolution operation and second convolution operation on the first vehicle image respectively to obtain first convolution feature maps and second convolution feature maps with different receptive fields, wherein the size of the convolution kernel of the first convolution operation is smaller than the size of the convolution kernel of the second convolution operation; fusing the first convolution feature maps and the second convolution feature maps to obtain target convolution feature maps; and performing license plate recognition based on the target convolution feature maps to obtain a target license plate recognition result. The application can describe objects of different sizes with different receptive fields, so that features on the first vehicle image can be detected and extracted from different sizes, and the feature representation capability is stronger, so that the accuracy of license plate recognition can be improved.

Description

Technical Field

[0001] This application mainly relates to the field of license plate recognition technology, specifically to a license plate recognition method and device. Background Technology

[0002] With the advancement of technology, cars are becoming increasingly common in daily life. License plates are unique identification codes for vehicles, and efficient license plate recognition has wide applications in traffic management, park management, and vehicle dispatching. Using computer vision for license plate recognition can save significant manpower and resources while maintaining higher accuracy. For truck dispatching tasks in smart logistics parks and smart transportation systems, cameras often employ suspended far-field shooting. Due to the unique characteristics of these scenarios, existing visual algorithms cannot achieve robust recognition in complex environments. The main reasons include: vehicle motion blur, license plate tilt, small license plate area, occlusion, and environmental factors such as lighting, rain, and snow. Conventional neural network models have weak feature extraction capabilities, resulting in low accuracy in license plate recognition.

[0003] In other words, the accuracy of license plate recognition in existing technologies is not high. Summary of the Invention

[0004] This application provides a license plate recognition method and apparatus, aiming to solve the problem of low accuracy in license plate recognition in the prior art.

[0005] Firstly, this application provides a license plate recognition method, the license plate recognition method comprising:

[0006] Acquire the image of the first vehicle;

[0007] The first vehicle image is subjected to a first convolution operation and a second convolution operation respectively to obtain a first convolution feature map and a second convolution feature map with different receptive fields, wherein the size of the convolution kernel of the first convolution operation is smaller than the size of the convolution kernel of the second convolution operation.

[0008] The first convolutional feature map and the second convolutional feature map are fused to obtain the target convolutional feature map;

[0009] License plate recognition is performed based on the target convolutional feature map to obtain the target license plate recognition result.

[0010] Optionally, fusing the first convolutional feature map and the second convolutional feature map to obtain the target convolutional feature map includes:

[0011] A third convolution operation is performed on the first vehicle image to obtain a third convolution feature map. The kernel size of the third convolution operation is the same as that of the second convolution operation. The third convolution operation and the second convolution operation are dilated convolutions. The dilation rates of the third convolution operation and the second convolution operation are different.

[0012] The target convolutional feature map is obtained by fusing the first convolutional feature map, the second convolutional feature map, and the third convolutional feature map.

[0013] Optionally, the step of performing license plate recognition based on the target convolutional feature map to obtain the target license plate recognition result includes:

[0014] Detect vehicle region images on the first vehicle image based on the target convolutional feature map;

[0015] Perform rectangular license plate region detection on the vehicle region image to obtain a rectangular license plate region image;

[0016] The rectangular license plate area image is subjected to text recognition to obtain the license plate text recognition result;

[0017] The target license plate recognition result is determined based on the license plate text recognition result.

[0018] Optionally, the step of performing rectangular license plate region detection on the vehicle region image to obtain a rectangular license plate region image includes:

[0019] The vehicle region image is input into a first feature extraction layer and a second feature extraction layer to obtain a first vehicle feature map and a second vehicle feature map. The first vehicle feature map is obtained by extracting features from the vehicle region image through the first feature extraction layer, and the second vehicle feature map is obtained by extracting features from the vehicle region image through the first feature extraction layer and the second feature extraction layer in sequence.

[0020] The second vehicle feature map and the first vehicle feature map are upsampled and fused to obtain the third vehicle feature map;

[0021] The third vehicle feature map and the second vehicle feature map are downsampled and fused to obtain the fourth vehicle feature map;

[0022] Based on the fourth vehicle feature map, a rectangular license plate region detection is performed to obtain a rectangular license plate region image.

[0023] Optionally, the step of detecting a rectangular license plate region based on the fourth vehicle feature map to obtain a rectangular license plate region image includes:

[0024] Based on the fourth vehicle feature map, a distorted license plate region is detected to obtain the distorted license plate region;

[0025] Based on the fourth vehicle feature map, affine parameter regression is performed to obtain the license plate affine parameters;

[0026] The distorted license plate region is corrected based on the affine parameters of the license plate to obtain the rectangular license plate region image.

[0027] Optionally, acquiring the first vehicle image includes:

[0028] Acquire at least two consecutive images of the second vehicle captured by a camera device as the vehicle enters or exits the scene;

[0029] Perform a difference operation on the at least two second vehicle images to obtain a difference image;

[0030] Determine whether the area ratio of the moving object region on the difference image falls within a preset ratio range;

[0031] If the area ratio of the moving object region on the differential image is within a preset ratio range, then the last image among the at least two second vehicle images is determined as the first vehicle image.

[0032] Optionally, determining the target license plate recognition result based on the license plate text recognition result includes:

[0033] The capture time of the first vehicle image and the transit time of multiple vehicles entering and exiting the scene are obtained.

[0034] Multiple target license plate numbers are determined based on the shooting time, wherein the multiple target license plate numbers are the license plate numbers of vehicles whose elapsed time and shooting time do not differ from the preset time.

[0035] Determine whether there is a target license plate number among the plurality of target license plate numbers that has a similarity higher than a preset similarity with the license plate text recognition result;

[0036] If it exists, the target license plate number with the highest similarity to the license plate text recognition result among the multiple target license plate numbers is determined as the target license plate recognition result.

[0037] Secondly, this application provides a license plate recognition device, the license plate recognition device comprising:

[0038] The first acquisition unit is used to acquire a first vehicle image;

[0039] The second convolution operation unit is used to perform a first convolution operation and a second convolution operation on the first vehicle image respectively to obtain a first convolution feature map and a second convolution feature map with different receptive fields, wherein the size of the convolution kernel of the first convolution operation is smaller than the size of the convolution kernel of the second convolution operation.

[0040] A fusion unit is used to fuse the first convolutional feature map and the second convolutional feature map to obtain a target convolutional feature map;

[0041] The license plate recognition unit is used to perform license plate recognition based on the target convolutional feature map to obtain the target license plate recognition result.

[0042] Optionally, the fusion unit is used for:

[0043] A third convolution operation is performed on the first vehicle image to obtain a third convolution feature map. The kernel size of the third convolution operation is the same as that of the second convolution operation. The third convolution operation and the second convolution operation are dilated convolutions. The dilation rates of the third convolution operation and the second convolution operation are different.

[0044] The target convolutional feature map is obtained by fusing the first convolutional feature map, the second convolutional feature map, and the third convolutional feature map.

[0045] Optionally, the license plate recognition unit is used for:

[0046] Detect vehicle region images on the first vehicle image based on the target convolutional feature map;

[0047] Perform rectangular license plate region detection on the vehicle region image to obtain a rectangular license plate region image;

[0048] The rectangular license plate area image is subjected to text recognition to obtain the license plate text recognition result;

[0049] The target license plate recognition result is determined based on the license plate text recognition result.

[0050] Optionally, the license plate recognition unit is used for:

[0051] The vehicle region image is input into a first feature extraction layer and a second feature extraction layer to obtain a first vehicle feature map and a second vehicle feature map. The first vehicle feature map is obtained by extracting features from the vehicle region image through the first feature extraction layer, and the second vehicle feature map is obtained by extracting features from the vehicle region image through the first feature extraction layer and the second feature extraction layer in sequence.

[0052] The second vehicle feature map and the first vehicle feature map are upsampled and fused to obtain the third vehicle feature map;

[0053] The third vehicle feature map and the second vehicle feature map are downsampled and fused to obtain the fourth vehicle feature map;

[0054] Based on the fourth vehicle feature map, a rectangular license plate region detection is performed to obtain a rectangular license plate region image.

[0055] Optionally, the license plate recognition unit is used for:

[0056] Based on the fourth vehicle feature map, a distorted license plate region is detected to obtain the distorted license plate region;

[0057] Based on the fourth vehicle feature map, affine parameter regression is performed to obtain the license plate affine parameters;

[0058] The distorted license plate region is corrected based on the affine parameters of the license plate to obtain the rectangular license plate region image.

[0059] Optionally, the first acquisition unit is configured to:

[0060] Acquire at least two consecutive images of the second vehicle captured by a camera device as the vehicle enters or exits the scene;

[0061] Perform a difference operation on the at least two second vehicle images to obtain a difference image;

[0062] Determine whether the area ratio of the moving object region on the difference image falls within a preset ratio range;

[0063] If the area ratio of the moving object region on the differential image is within a preset ratio range, then the last image among the at least two second vehicle images is determined as the first vehicle image.

[0064] Optionally, the license plate recognition unit is used for:

[0065] The capture time of the first vehicle image and the transit time of multiple vehicles entering and exiting the scene are obtained.

[0066] Multiple target license plate numbers are determined based on the shooting time, wherein the multiple target license plate numbers are the license plate numbers of vehicles whose elapsed time and shooting time do not differ from the preset time.

[0067] Determine whether there is a target license plate number among the plurality of target license plate numbers that has a similarity higher than a preset similarity with the license plate text recognition result;

[0068] If it exists, the target license plate number with the highest similarity to the license plate text recognition result among the multiple target license plate numbers is determined as the target license plate recognition result.

[0069] Thirdly, this application provides a computer device, the computer device comprising:

[0070] One or more processors;

[0071] Memory; and

[0072] One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the license plate recognition method as described in any of the first aspects.

[0073] Fourthly, this application provides a computer-readable storage medium storing a plurality of instructions adapted for loading by a processor to perform the steps of the license plate recognition method described in any one of the first aspects.

[0074] This application provides a license plate recognition method and apparatus. The license plate recognition method includes: acquiring a first vehicle image; performing a first convolution operation and a second convolution operation on the first vehicle image respectively to obtain a first convolution feature map and a second convolution feature map with different receptive fields, wherein the size of the convolution kernel of the first convolution operation is smaller than the size of the convolution kernel of the second convolution operation; fusing the first convolution feature map and the second convolution feature map to obtain a target convolution feature map; and performing license plate recognition based on the target convolution feature map to obtain a target license plate recognition result. This application, when acquiring a first vehicle image, uses convolution operations with convolution kernels of different sizes to extract features from the first vehicle image, thereby obtaining a first convolution feature map and a second convolution feature map with different receptive fields. Then, it fuses the first convolution feature map and the second convolution feature map for license plate detection. Compared with traditional detection methods, different receptive fields can be used to describe objects of different sizes. Therefore, features on the first vehicle image can be detected and extracted from different perspectives, resulting in stronger feature representation capabilities and thus improving the accuracy of license plate recognition. Attached Figure Description

[0075] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0076] Figure 1 This is a schematic diagram of a license plate recognition system provided in an embodiment of this application.

[0077] Figure 2 This is a schematic flowchart of an embodiment of the license plate recognition method provided in this application.

[0078] Figure 3 This is a schematic diagram of the process of detecting a rectangular license plate region in a vehicle region image to obtain a rectangular license plate region image in one embodiment of the license plate recognition method provided in this application.

[0079] Figure 4 This is a schematic diagram of the network structure of an embodiment of the license plate detection model in this application.

[0080] Figure 5 This is a schematic diagram of an embodiment of the license plate recognition device provided in this application.

[0081] Figure 6 This is a schematic diagram of an embodiment of the computer device provided in this application. Detailed Implementation

[0082] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0083] In the description of this application, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more features. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

[0084] In this application, the term "exemplary" is used to mean "used as an example, illustration, or description." Any embodiment described as "exemplary" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes are not described in detail to avoid obscuring the description of this application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0085] This application provides a license plate recognition method and apparatus, which will be described in detail below.

[0086] Please see Figure 1 , Figure 1 This is a schematic diagram of a license plate recognition system provided in an embodiment of this application. The license plate recognition system may include a computer device 100, which integrates a license plate recognition device.

[0087] In this embodiment, the computer device 100 can be a standalone server, a server network, or a server cluster. For example, the computer device 100 described in this embodiment includes, but is not limited to, a computer, a network host, a single network server, a set of multiple network servers, or a cloud server composed of multiple servers. The cloud server is composed of a large number of computers or network servers based on cloud computing.

[0088] In this embodiment, the computer device 100 described above can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device 100 can be a desktop computer, a portable computer, a network server, a handheld computer (Personal Digital Assistant, PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, an embedded device, etc. This embodiment does not limit the type of computer device 100.

[0089] Those skilled in the art will understand that Figure 1 The application environment shown is merely one application scenario of the solution in this application and does not constitute a limitation on the application scenario of the solution in this application. Other application environments may include more than one application scenario. Figure 1 The number of computer devices shown is more or less, for example Figure 1Only one computer device is shown in the image. It is understood that the license plate recognition system may also include one or more other computer devices capable of processing data, which are not specifically limited here.

[0090] In addition, such as Figure 1 As shown, the license plate recognition system may also include a memory 200 for storing data.

[0091] It should be noted that, Figure 1 The schematic diagram of the license plate recognition system shown is merely an example. The license plate recognition system and scenarios described in this application are intended to more clearly illustrate the technical solutions of this application and do not constitute a limitation on the technical solutions provided in this application. As those skilled in the art will know, with the evolution of license plate recognition systems and the emergence of new business scenarios, the technical solutions provided in this application are also applicable to similar technical problems.

[0092] First, this application provides a license plate recognition method, which includes: acquiring a first vehicle image; performing a first convolution operation and a second convolution operation on the first vehicle image respectively to obtain a first convolution feature map and a second convolution feature map with different receptive fields, wherein the size of the convolution kernel of the first convolution operation is smaller than the size of the convolution kernel of the second convolution operation; fusing the first convolution feature map and the second convolution feature map to obtain a target convolution feature map; and performing license plate recognition based on the target convolution feature map to obtain a target license plate recognition result. This application uses convolution operations with convolution kernels of different sizes to extract features from the first vehicle image when acquiring it, thereby obtaining a first convolution feature map and a second convolution feature map with different receptive fields. Then, it fuses the first convolution feature map and the second convolution feature map for license plate detection. Compared with traditional detection methods, different receptive fields can be used to describe objects of different sizes. Therefore, features on the first vehicle image can be detected and extracted from different perspectives, resulting in stronger feature representation capabilities and thus improving the accuracy of license plate recognition.

[0093] like Figure 2 As shown, Figure 2 This is a schematic flowchart of an embodiment of the license plate recognition method provided in this application, which includes the following steps S201 to S204:

[0094] S201. Obtain the image of the first vehicle.

[0095] The first vehicle image can be an image captured by a camera device of a vehicle entering or exiting a scene. This scene can be a parking lot, a platform, or similar location. For example, in a platform scene, the camera device is a webcam, positioned 3-5 meters above the platform and approximately 2-8 meters from the platform boundary. The camera is installed using a top-mounted method. There are no strict requirements on the specific model of the camera, but it must have a resolution of 200W or 400W, support RTSP / RTMP protocols, and have a 2.8-12mm zoom lens to meet the installation needs of different areas.

[0096] The camera of this invention is connected to a computer device, which is an edge server. The server is responsible for deploying algorithms and model inference. The hardware platform used by the edge server is an Intel(R) Xeon(R) CPU E5-26xx v4, 48GB RAM, and dual Tesla T4 graphics cards. The software platform is an Ubuntu 16.04.4 operating system, CUDA 10.0, cuDNN 7.6, and the Torch 1.8.1 deep learning framework, which is deployed using Docker containers.

[0097] To avoid the computational burden of detecting every image from the camera, in one specific embodiment, acquiring the first vehicle image may include:

[0098] (1) Acquire at least two consecutive images of the second vehicle obtained by the camera device capturing the scene of the vehicle entering and exiting.

[0099] In this embodiment, images captured by the camera are acquired at a preset period to obtain at least two consecutive images of the second vehicle. The preset period can be 0.1 seconds, etc., and can be set according to specific circumstances.

[0100] (2) Perform a difference operation on at least two second vehicle images to obtain a difference image.

[0101] In one specific embodiment, the image frame difference method is used to perform differential operations on at least two second vehicle images to obtain a difference image. The image frame difference method is one of the most commonly used methods for moving target detection and segmentation. Its basic principle is to extract moving regions from an image by using pixel-based temporal difference between two or three adjacent frames and then applying thresholding. First, the corresponding pixel values ​​of adjacent frames are subtracted to obtain the difference image. Then, the difference image is binarized. If the change in corresponding pixel value is less than a predetermined threshold when the ambient brightness change is small, this area can be considered a background pixel. If the change in pixel value of an image region is not less than a predetermined threshold, it can be considered to be caused by a moving object in the image. These regions are marked as foreground pixels. The marked pixel regions can be used to determine the position of the moving target in the image. Because the time interval between two adjacent frames is very short, using the previous frame as the background model for the current frame has good real-time performance, does not accumulate background, and has a fast update speed, simple algorithm, and low computational cost. The algorithm's weakness lies in its sensitivity to environmental noise. The choice of threshold is crucial; too low a threshold is insufficient to suppress noise in the image, while too high a threshold ignores useful variations. For large, uniformly colored moving targets, it may create holes within the target, making it impossible to extract the moving target completely.

[0102] (3) Determine whether the area ratio of the moving object region on the difference image belongs to the preset ratio range.

[0103] In one specific embodiment, the preset proportion range can be set according to specific circumstances. In one specific embodiment, multiple pre-captured vehicle images captured by a camera device are acquired in advance. The proportion of the labeled vehicle area on each manually labeled pre-captured vehicle image to the labeled area of ​​the pre-captured vehicle image is obtained. The expected value and variance of multiple labeled area proportions are obtained. The difference between the expected value and twice the variance is determined as the minimum value of the preset proportion range, and the sum of the expected value and twice the variance is determined as the maximum value of the preset proportion range.

[0104] (4) If the area ratio of the moving object region on the differential image is within the preset ratio range, then the last image among at least two second vehicle images shall be determined as the first vehicle image.

[0105] If the area of ​​the moving object region in the difference image falls within a preset range, it indicates that the moving object region may be a vehicle. In this case, the last image among at least two second vehicle images is designated as the first vehicle image. By detecting moving objects, images containing moving objects can be retained as the first vehicle images for license plate detection, thus avoiding the need to detect license plates in every image.

[0106] S202. Perform a first convolution operation and a second convolution operation on the first vehicle image respectively to obtain a first convolution feature map and a second convolution feature map with different receptive fields.

[0107] In this embodiment, the receptive field in a convolutional neural network is defined as the size of the region on the input image that corresponds to a pixel on the feature map output by each layer of the convolutional neural network. Specifically, one point on the feature map corresponds to a region on the input image. In this embodiment, the receptive fields of the first convolutional feature map and the second convolutional feature map are different.

[0108] In one specific embodiment, the kernel size of the first convolution operation is smaller than the kernel size of the second convolution operation. Specifically, the first convolution operation uses a 1×1 convolution, and the kernel size is 1*1. The first convolution operation is a standard convolution to preserve the original receptive field. Alternatively, the first convolution operation can use a 3×3 convolution, and the kernel size used in the second convolution operation is 3*3.

[0109] In this embodiment of the application, for example, feature extraction is performed on the first vehicle image to obtain a 256-dimensional feature map. The 256-dimensional feature map is then subjected to a first convolution operation to obtain a 1x1x256 first convolutional feature map. The 256-dimensional feature map is then subjected to a second convolution operation to obtain a 9x1x256 second convolutional feature map.

[0110] S203. The first convolutional feature map and the second convolutional feature map are fused to obtain the target convolutional feature map.

[0111] Specifically, the first convolutional feature map (1x1x256) and the second convolutional feature map (9x1x256) are superimposed along the channel dimension to obtain the target convolutional feature map.

[0112] To increase the receptive field and improve feature extraction capability, in a specific embodiment, the first convolutional feature map and the second convolutional feature map are fused to obtain the target convolutional feature map, which may include:

[0113] (1) Perform a third convolution operation on the first vehicle image to obtain the third convolution feature map.

[0114] The kernel size of the third convolution operation is the same as that of the second convolution operation. The third convolution operation and the second convolution operation are dilated convolutions. The dilation rates of the third convolution operation and the second convolution operation are different.

[0115] Traditional deep convolutional neural networks (CNNs) typically perform convolution on the image for feature extraction and dimensionality reduction before pooling to reduce its size. As the network depth increases, pooling layers cause the image size to shrink further. When upsampling is needed to enlarge the image to its original size, this leads to the loss of internal data structure, spatial hierarchy information, and information related to the reconstruction of small objects, potentially preventing significant improvements in network accuracy. Dilated convolution operates on the same principle as regular convolution, but it introduces a new parameter called the "dilation rate." As the name suggests, the dilation rate represents the size of the kernel expansion, i.e., the distance between the parameters within the kernel. Regular convolution has a dilation rate of 1, meaning the kernel does not expand. Dilated convolution, without pooling layers, increases the receptive field, ensuring each convolution output contains a larger range of information, thus improving network performance.

[0116] In one specific embodiment, the third convolution operation and the second convolution operation are dilated convolutions, with the second convolution operation having a dilation rate of 1 and the third convolution operation having a dilation rate of 2. Alternatively, the second convolution operation could have a dilation rate of 2 and the third convolution operation could have a dilation rate of 4.

[0117] (2) The first convolutional feature map, the second convolutional feature map and the third convolutional feature map are fused to obtain the target convolutional feature map.

[0118] Specifically, the first, second, and third convolutional feature maps are superimposed and fused along the channel dimension to obtain the target convolutional feature map.

[0119] In other embodiments, a fourth convolution operation can be performed on the first vehicle image to obtain a fourth convolutional feature map. The kernel sizes of the third, second, and fourth convolution operations are the same, and these operations are dilated convolutions with different dilation rates. For example, the dilation rate of the second convolution operation is 1, the dilation rate of the third convolution operation is 2, and the dilation rate of the fourth convolution operation is 4. By performing dilated convolutions on three of these branches, a feature map with an expanded receptive field is obtained. Fusing the feature maps from all four branches results in a stronger representation capability that better reflects the characteristics of roads and vehicles on ramps.

[0120] S204. Based on the target convolutional feature map, perform license plate recognition to obtain the target license plate recognition result.

[0121] In one specific embodiment, license plate recognition is performed based on the target convolutional feature map to obtain the target license plate recognition result, including:

[0122] (1) Detect vehicle region images on the first vehicle image based on target convolutional feature maps.

[0123] In one specific embodiment, the target convolutional feature map is input into a first object detection model to obtain a vehicle detection box, and the image within the vehicle detection box is defined as the vehicle region image. The first object detection model can be YOLOv5, YOLOv4, etc. The vehicle region image is the image within the bounding rectangle of the vehicle.

[0124] In one specific embodiment, the first vehicle image is scaled to a preset size, specifically 416*256. This application addresses the scenario of freight vehicle license plate recognition in a logistics park. Due to the requirement for uniformity of equipment within the park, images are scaled to a fixed size. The standard image input size for YOLOv5 is 416*416. Since this invention uses a uniform input size (standard input is 1920*1080) and does not involve other input sizes, the image adaptive scaling module is removed. Simultaneously, to adapt to the network structure input, this invention calculates an adaptive size. This invention uses a length scaling ratio of 0.2166 and a width scaling ratio of 0.385. Using the smaller ratio, the scaled length is calculated to be 416 and the width to be 234. Mod(416-234, 32) = 22. Because it undergoes five subsampling iterations, taking the remainder of 2 to the power of 5 gives a width padded with 22 pixels, resulting in a final size of 416*256. By using a uniform image input size, the time consumption associated with adaptive size calculation is reduced.

[0125] (2) Perform rectangular license plate region detection on the vehicle region image to obtain a rectangular license plate region image.

[0126] In one specific embodiment, the target convolutional feature map is input into a second object detection model to obtain a license plate detection box. The image within the license plate detection box is then defined as a rectangular license plate region image. The second object detection model can be YOLOv5, YOLOv4, etc. The rectangular license plate region image is the image within the bounding rectangle of the license plate.

[0127] (3) Perform text recognition on the rectangular license plate area image to obtain the license plate text recognition result.

[0128] In one specific embodiment, CRNN is used to perform text recognition on a rectangular license plate area image to obtain the license plate text recognition result.

[0129] This application first detects the vehicle, then detects the license plate, and finally performs text recognition on the detected license plate, which can improve the accuracy of license plate recognition.

[0130] (4) Determine the target license plate recognition result based on the license plate text recognition result.

[0131] In one specific embodiment, the license plate text recognition result is determined as the target license plate recognition result. The target license plate recognition result is the license plate number of the vehicle in the first vehicle image.

[0132] To improve recognition accuracy, in one specific embodiment, the target license plate recognition result is determined based on the license plate text recognition result, including:

[0133] (1) Obtain the capture time of the first vehicle image and the transit time of multiple vehicles entering and exiting the scene.

[0134] Specifically, the arrival and departure times of multiple vehicles in the vehicle entry and exit scenarios are predicted in advance by the system. In the smart park vehicle scheduling scenario, the gate can obtain the license plate number of the vehicle to be entering the port and the departure and departure times of the vehicle.

[0135] (2) Determine multiple target license plate numbers based on the shooting time. Among them, multiple target license plate numbers are the license plate numbers of vehicles whose time difference from shooting time does not exceed a preset time.

[0136] The preset time can be 5 minutes, but can be set according to specific circumstances. If the passing times of multiple target license plate numbers are close to the shooting time, these license plate numbers are highly likely to include the license plate in the first vehicle image.

[0137] (3) Determine whether there is a target license plate number among multiple target license plate numbers that has a higher similarity to the license plate text recognition result than the preset similarity.

[0138] The preset similarity can be 0.8, but can be set according to specific requirements. Specifically, the TF-IDF algorithm is used to calculate the similarity between the license plate text recognition result and multiple target license plate numbers, obtaining the similarity between the license plate text recognition result and each target license plate number, and determining whether the similarity between the license plate text recognition result and each target license plate number is higher than the preset similarity.

[0139] (3) If it exists, the target license plate number with the highest similarity to the license plate text recognition result among multiple target license plate numbers shall be determined as the target license plate recognition result.

[0140] This invention draws on text matching algorithms from natural language processing, using the TF-IDF algorithm to calculate similarity. Target license plate numbers with a similarity greater than 0.8 to the recognized license plate text are identified as the target license plate recognition result, further improving accuracy.

[0141] To improve license plate feature extraction capabilities, please refer to... Figure 3 and Figure 4 In one specific embodiment, performing rectangular license plate region detection on the vehicle region image to obtain a rectangular license plate region image may include:

[0142] S301. Input the vehicle region image into the first feature extraction layer and the second feature extraction layer to obtain the first vehicle feature map and the second vehicle feature map.

[0143] In one specific embodiment, a preset training set is obtained, and a license plate detection model is trained based on the preset training set. The license plate detection model includes a first feature extraction layer and a second feature extraction layer. The vehicle region image is input into the first feature extraction layer and the second feature extraction layer of the license plate detection model to obtain a first vehicle feature map and a second vehicle feature map.

[0144] The first vehicle feature map is obtained by extracting features from the license plate image to be identified through the first feature extraction layer, and the second vehicle feature map is obtained by extracting features from the license plate image to be identified through the first and second feature extraction layers in sequence. The size of the first vehicle feature map is smaller than the size of the second vehicle feature map.

[0145] In a specific embodiment, the first feature extraction layer can be conv3*3 32+resblock(32)+MaxPooling. The second feature extraction layer can be conv3*3 64+resblock(64)+MaxPooling.

[0146] Of course, a license plate detection model can include a first detection layer and multiple second detection layers.

[0147] Furthermore, the first feature extraction layer can be conv3*3 32+CBAM+resblock(32)+MaxPooling. The second feature extraction layer can be conv3*3 64+CBAM+resblock(64)+MaxPooling. The Convolutional Block Attention Module (CBAM) represents the attention mechanism module of the convolutional module, which is a kind of attention mechanism module that combines spatial and channel attention. Compared with the attention mechanism of SENet that only focuses on channels, it can achieve better results.

[0148] S302. Upsample and fuse the second vehicle feature map and the first vehicle feature map to obtain the third vehicle feature map.

[0149] In one specific embodiment, the second vehicle feature map is upsampled by a factor of two to obtain an upsampled second vehicle feature map. The upsampled second vehicle feature map and the first vehicle feature map are then fused along the channel dimension to obtain a third vehicle feature map. The main purpose of upsampling is to enlarge the original image so that it can be displayed on a higher resolution display device.

[0150] S303. Downsample and fuse the third vehicle feature map and the second vehicle feature map to obtain the fourth vehicle feature map.

[0151] In one specific embodiment, the third vehicle feature map is downsampled by a factor of two to obtain a downsampled third vehicle feature map. The downsampled third vehicle feature map and the second vehicle feature map are then fused along the channel dimension to obtain a fourth vehicle feature map. Downsampling is primarily used to make the image conform to the size of the display area and to generate a thumbnail of the corresponding image.

[0152] S304. Based on the fourth vehicle feature map, a rectangular license plate region detection is performed to obtain a rectangular license plate region image.

[0153] Considering that each vehicle has only one license plate, but the position of the license plate in the vehicle image varies due to different shooting angles within the park, this invention does not aim to obtain a network model with strong spatial distribution through network training. This invention uses pyramid feature extraction to enhance the feature representation capability of the license plate detection model. Using this network feature enhancement strategy can improve the accuracy of license plate detection and localization.

[0154] In one specific embodiment, the fourth vehicle feature map is input into a third object detection model to detect a rectangular license plate region, resulting in a rectangular license plate region image. The third object detection model can be YOLOv5, YOLOv4, etc. The rectangular license plate region image is the image within the bounded rectangle of the license plate.

[0155] In another specific embodiment, detecting a rectangular license plate region based on the fourth vehicle feature map to obtain a rectangular license plate region image may include:

[0156] (1) Based on the fourth vehicle feature map, the distorted license plate region is detected to obtain the distorted license plate region.

[0157] In one specific embodiment, SOFTMAX regression is used to detect the distorted license plate region on the fourth vehicle feature map input to obtain the distorted license plate region. Specifically, the fourth vehicle feature map is convolved using Conv(3*3,2) and then subjected to SOFTMAX regression to obtain the distorted license plate region.

[0158] (2) Based on the fourth vehicle feature map, perform affine parameter regression to obtain the license plate affine parameters.

[0159] In one specific embodiment, affine parameter regression is performed on the fourth vehicle feature map to obtain the license plate affine parameters. Specifically, the fourth vehicle feature map is convolved using Conv(3*3,6) and then subjected to LINEAR regression to obtain the distorted license plate region.

[0160] In one specific embodiment, the license plate detection model includes a first detection layer and a second detection layer. The first detection layer includes a convolutional layer Conv(3*3,2) and a SOFTMAX regression layer. The second detection layer includes a convolutional layer Conv(3*3,6) and a LINEAR linear regression layer.

[0161] (3) The distorted license plate region is corrected based on the license plate affine parameters to obtain a rectangular license plate region image.

[0162] Furthermore, a preset training set is obtained, and a license plate detection model is trained based on the preset training set. This includes: obtaining an initial training set, performing data augmentation on the initial training set to obtain a data-augmented preset training set, and training the license plate detection model based on the preset training set. The initial training set includes multiple labeled vehicle region images with license plate areas. Data augmentation of the initial training set can be performed using Mosiac, MixUp, and CutMix. This invention aims to improve the robustness of the algorithm in the current scenario, enabling effective detection and recognition of license plates when new vehicles enter the port. This invention introduces three data augmentation methods—Mosiac, MixUp, and CutMix—to expand the data, hoping to change the data distribution and increase the amount of data so that the network can better fit the data. In addition to the above data augmentation, random perturbations of color space brightness and saturation are used to simulate the complex environment of real-world scenes, including self-augmentation strategies such as Mosaic, rotation, flipping, scaling, and white balance.

[0163] Furthermore, addressing the diversity of Chinese license plates and the diversity of license plates for trucks and port vehicles, this invention uses a perspective-transformed Chinese license plate. Obtaining the initial training set includes: randomly generating license plate content; generating license plate text based on the font and adding license plate backgrounds of various colors; randomly generating affine transformations of the license plate; randomly adding noise with saturated lighting; adding Gaussian blur and Gaussian features; and pasting the perspective-transformed license plate onto a real vehicle image to obtain an image of the vehicle region marked with the license plate area. Through these six steps, this invention can generate a large number of random license plate images and image of the vehicle region marked with the license plate area. By performing data augmentation on these images and training them, a more robust license plate detection model can be obtained.

[0164] To better implement the license plate recognition method in this application embodiment, based on the license plate recognition method, this application embodiment also provides a license plate recognition device, such as... Figure 5 As shown, the license plate recognition device 500 includes:

[0165] The first acquisition unit 501 is used to acquire a first vehicle image;

[0166] The second convolution operation unit 502 is used to perform a first convolution operation and a second convolution operation on the first vehicle image respectively to obtain a first convolution feature map and a second convolution feature map with different receptive fields, wherein the size of the convolution kernel of the first convolution operation is smaller than the size of the convolution kernel of the second convolution operation.

[0167] The fusion unit 503 is used to fuse the first convolutional feature map and the second convolutional feature map to obtain the target convolutional feature map.

[0168] The license plate recognition unit 504 is used to perform license plate recognition based on the target convolutional feature map to obtain the target license plate recognition result.

[0169] Optionally, the fusion unit 503 is used for:

[0170] A third convolution operation is performed on the first vehicle image to obtain a third convolution feature map. The kernel size of the third convolution operation is the same as that of the second convolution operation. The third convolution operation and the second convolution operation are dilated convolutions. The dilation rates of the third convolution operation and the second convolution operation are different.

[0171] The target convolutional feature map is obtained by fusing the first, second, and third convolutional feature maps.

[0172] Optionally, the license plate recognition unit 504 is used for:

[0173] Vehicle region image detection on the first vehicle image based on target convolutional feature map;

[0174] Rectangular license plate region detection is performed on the vehicle region image to obtain a rectangular license plate region image;

[0175] Perform text recognition on a rectangular license plate area image to obtain the license plate text recognition results;

[0176] The target license plate recognition result is determined based on the license plate text recognition result.

[0177] Optionally, the license plate recognition unit 504 is used for:

[0178] The vehicle region image is input into the first feature extraction layer and the second feature extraction layer to obtain the first vehicle feature map and the second vehicle feature map. The first vehicle feature map is obtained by extracting features from the vehicle region image through the first feature extraction layer, and the second vehicle feature map is obtained by extracting features from the vehicle region image through the first feature extraction layer and the second feature extraction layer in sequence.

[0179] The second vehicle feature map and the first vehicle feature map are upsampled and fused to obtain the third vehicle feature map;

[0180] The third vehicle feature map and the second vehicle feature map are downsampled and fused to obtain the fourth vehicle feature map;

[0181] Rectangular license plate region detection is performed based on the fourth vehicle feature map to obtain a rectangular license plate region image.

[0182] Optionally, the license plate recognition unit 504 is used for:

[0183] The distorted license plate region is detected based on the fourth vehicle feature map.

[0184] Affine parameter regression is performed based on the fourth vehicle feature map to obtain the license plate affine parameters;

[0185] The distorted license plate region is corrected based on the affine parameters of the license plate to obtain a rectangular license plate region image.

[0186] Optionally, the first acquisition unit 501 is used for:

[0187] Acquire at least two consecutive images of the second vehicle captured by a camera device as the vehicle enters or exits the scene;

[0188] Perform a difference operation on at least two images of the second vehicle to obtain a difference image;

[0189] Determine whether the area ratio of the moving object region on the difference image falls within a preset ratio range;

[0190] If the area ratio of the moving object region on the differential image is within a preset ratio range, then the last image among at least two second vehicle images will be determined as the first vehicle image.

[0191] Optionally, the license plate recognition unit 504 is used for:

[0192] The capture time of the first vehicle image and the transit time of multiple vehicles entering and exiting the scene are obtained.

[0193] Multiple target license plate numbers are determined based on the shooting time. Among them, the multiple target license plate numbers are the license plate numbers of vehicles whose time difference between the shooting time and the time of shooting does not exceed a preset time.

[0194] Determine whether there is a target license plate number among multiple target license plate numbers that has a similarity to the license plate text recognition result that is higher than a preset similarity;

[0195] If it exists, the target license plate number with the highest similarity to the license plate text recognition result among multiple target license plate numbers will be determined as the target license plate recognition result.

[0196] This application also provides a computer device that integrates any of the license plate recognition devices provided in this application. The computer device includes:

[0197] One or more processors;

[0198] Memory; and

[0199] One or more applications, wherein the applications are stored in memory and configured to be executed by a processor as steps of the license plate recognition method in any of the embodiments described above.

[0200] like Figure 6 As shown, it illustrates a structural schematic diagram of the computer device involved in the embodiments of this application, specifically:

[0201] The computer device may include components such as a processor 601 with one or more processing cores, a memory 602 with one or more computer-readable storage media, a power supply 603, and an input unit 604. Those skilled in the art will understand that the computer device structure shown in the figures does not constitute a limitation on the computer device, and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein:

[0202] Processor 601 is the control center of the computer device. It connects various parts of the computer device via various interfaces and lines, and performs various functions and processes data by running or executing software programs and / or modules stored in memory 602, and by calling data stored in memory 602, thereby providing overall monitoring of the computer device. Optionally, processor 601 may include one or more processing cores; processor 601 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. Preferably, processor 601 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the aforementioned modem processor may not be integrated into processor 601.

[0203] The memory 602 can be used to store software programs and modules. The processor 601 executes various functional applications and data processing by running the software programs and modules stored in the memory 602. The memory 602 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 602 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 602 may also include a memory controller to provide the processor 601 with access to the memory 602.

[0204] The computer device also includes a power supply 603 that supplies power to the various components. Preferably, the power supply 603 can be logically connected to the processor 601 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 603 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0205] The computer device may also include an input unit 604, which can be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0206] Although not shown, the computer device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 601 in the computer device loads the executable files corresponding to the processes of one or more application programs into the memory 602 according to the following instructions, and the processor 601 runs the application programs stored in the memory 602 to realize various functions, as follows:

[0207] The process involves acquiring a first vehicle image; performing a first convolution operation and a second convolution operation on the first vehicle image to obtain first and second convolution feature maps with different receptive fields, wherein the size of the convolution kernel in the first convolution operation is smaller than the size of the convolution kernel in the second convolution operation; fusing the first and second convolution feature maps to obtain a target convolution feature map; and performing license plate recognition based on the target convolution feature map to obtain the target license plate recognition result. This application uses convolution operations with convolution kernels of different sizes to extract features from the first vehicle image upon acquisition, thereby obtaining first and second convolution feature maps with different receptive fields. Then, the first and second convolution feature maps are fused for license plate detection. Compared to traditional detection methods, different receptive fields can be used to describe objects of different sizes, thus allowing for the detection and extraction of features from different perspectives, resulting in stronger feature representation capabilities and improved license plate recognition accuracy.

[0208] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0209] Therefore, embodiments of this application provide a computer-readable storage medium, which may include: read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk, etc. A computer program is stored thereon, and the computer program is loaded by a processor to execute the steps in any of the license plate recognition methods provided in embodiments of this application. For example, the computer program loaded by the processor can execute the following steps:

[0210] The process involves acquiring a first vehicle image; performing a first convolution operation and a second convolution operation on the first vehicle image to obtain first and second convolution feature maps with different receptive fields, wherein the size of the convolution kernel in the first convolution operation is smaller than the size of the convolution kernel in the second convolution operation; fusing the first and second convolution feature maps to obtain a target convolution feature map; and performing license plate recognition based on the target convolution feature map to obtain the target license plate recognition result. This application uses convolution operations with convolution kernels of different sizes to extract features from the first vehicle image upon acquisition, thereby obtaining first and second convolution feature maps with different receptive fields. Then, the first and second convolution feature maps are fused for license plate detection. Compared to traditional detection methods, different receptive fields can be used to describe objects of different sizes, thus allowing for the detection and extraction of features from different perspectives, resulting in stronger feature representation capabilities and improved license plate recognition accuracy.

[0211] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the detailed descriptions of other embodiments above, which will not be repeated here.

[0212] In practice, each of the above units or structures can be implemented as an independent entity or can be arbitrarily combined to be implemented as the same or several entities. For the specific implementation of each of the above units or structures, please refer to the previous method embodiments, which will not be repeated here.

[0213] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0214] The license plate recognition method and apparatus provided in the embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A license plate recognition method, characterized in that, The license plate recognition method includes: Acquire at least two consecutive images of the second vehicle captured by a camera device as the vehicle enters or exits the scene; Perform a difference operation on the at least two second vehicle images to obtain a difference image; Determine whether the area ratio of the moving object region on the difference image falls within a preset ratio range; If the area ratio of the moving object region on the differential image is within a preset ratio range, then the last image among the at least two second vehicle images is determined as the first vehicle image. The first vehicle image is subjected to a first convolution operation and a second convolution operation respectively to obtain a first convolution feature map and a second convolution feature map with different receptive fields. The size of the convolution kernel of the first convolution operation is smaller than the size of the convolution kernel of the second convolution operation. The size of the convolution kernel of the first convolution operation is 1×1. The first convolutional feature map and the second convolutional feature map are superimposed on each other along the channel dimension, and the first convolutional feature map and the second convolutional feature map are fused together to obtain the target convolutional feature map; Detect vehicle region images on the first vehicle image based on the target convolutional feature map; Perform rectangular license plate region detection on the vehicle region image to obtain a rectangular license plate region image; The rectangular license plate area image is subjected to text recognition to obtain the license plate text recognition result; The target license plate recognition result is determined based on the license plate text recognition result.

2. The license plate recognition method according to claim 1, characterized in that, The process of fusing the first convolutional feature map and the second convolutional feature map to obtain the target convolutional feature map includes: A third convolution operation is performed on the first vehicle image to obtain a third convolution feature map. The kernel size of the third convolution operation is the same as that of the second convolution operation. The third convolution operation and the second convolution operation are dilated convolutions. The dilation rates of the third convolution operation and the second convolution operation are different. The target convolutional feature map is obtained by fusing the first convolutional feature map, the second convolutional feature map, and the third convolutional feature map.

3. The license plate recognition method according to claim 1, characterized in that, The step of performing rectangular license plate region detection on the vehicle region image to obtain a rectangular license plate region image includes: The vehicle region image is input into a first feature extraction layer and a second feature extraction layer to obtain a first vehicle feature map and a second vehicle feature map. The first vehicle feature map is obtained by extracting features from the vehicle region image through the first feature extraction layer, and the second vehicle feature map is obtained by extracting features from the vehicle region image through the first feature extraction layer and the second feature extraction layer in sequence. The second vehicle feature map and the first vehicle feature map are upsampled and fused to obtain the third vehicle feature map; The third vehicle feature map and the second vehicle feature map are downsampled and fused to obtain the fourth vehicle feature map; Based on the fourth vehicle feature map, a rectangular license plate region detection is performed to obtain a rectangular license plate region image.

4. The license plate recognition method according to claim 3, characterized in that, The process of detecting a rectangular license plate region based on the fourth vehicle feature map to obtain a rectangular license plate region image includes: Based on the fourth vehicle feature map, a distorted license plate region is detected to obtain the distorted license plate region; Based on the fourth vehicle feature map, affine parameter regression is performed to obtain the license plate affine parameters; The distorted license plate region is corrected based on the affine parameters of the license plate to obtain the rectangular license plate region image.

5. The license plate recognition method according to claim 1, characterized in that, The step of determining the target license plate recognition result based on the license plate text recognition result includes: The capture time of the first vehicle image and the transit time of multiple vehicles entering and exiting the scene are obtained. Multiple target license plate numbers are determined based on the shooting time, wherein the multiple target license plate numbers are the license plate numbers of vehicles whose elapsed time and shooting time do not differ from the preset time. Determine whether there is a target license plate number among the plurality of target license plate numbers that has a similarity higher than a preset similarity with the license plate text recognition result; If it exists, the target license plate number with the highest similarity to the license plate text recognition result among the multiple target license plate numbers is determined as the target license plate recognition result.

6. A license plate recognition device, characterized in that, The license plate recognition device includes: The first acquisition unit is used to acquire at least two consecutive second vehicle images captured by the camera device of the vehicle entering and exiting the scene; perform a difference operation on the at least two second vehicle images to obtain a difference image; determine whether the area ratio of the moving object region on the difference image is within a preset ratio range; if the area ratio of the moving object region on the difference image is within the preset ratio range, then determine the last image among the at least two second vehicle images as the first vehicle image. The second convolution operation unit is used to perform a first convolution operation and a second convolution operation on the first vehicle image respectively to obtain a first convolution feature map and a second convolution feature map with different receptive fields. The size of the convolution kernel of the first convolution operation is smaller than the size of the convolution kernel of the second convolution operation. The size of the convolution kernel of the first convolution operation is 1×1. The fusion unit is used to superimpose the first convolutional feature map and the second convolutional feature map in the channel dimension, and fuse the first convolutional feature map and the second convolutional feature map to obtain the target convolutional feature map; The license plate recognition unit is configured to detect a vehicle region image on the first vehicle image based on the target convolutional feature map; perform rectangular license plate region detection on the vehicle region image to obtain a rectangular license plate region image; perform text recognition on the rectangular license plate region image to obtain a license plate text recognition result; and determine the target license plate recognition result based on the license plate text recognition result.

7. A computer device, characterized in that, The computer device includes: One or more processors; Memory; and One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the license plate recognition method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, It stores a computer program, which is loaded by a processor to execute the steps of the license plate recognition method according to any one of claims 1 to 5.

Images