License plate selection method based on definition comparison

Abstract

The invention discloses a license plate selection method based on definition comparison, and relates to the technical field of image processing. The method comprises steps that a license plate image set is acquired, and the license plate image set is a group of multi-angle license plate images continuously acquired from the same vehicle in the same scene; performing definition parameter calculation based on pixel vector operation on the license plate image set one by one to determine a definition parameter set; the minimum value in the definition parameter set is taken, the license plate selection result is determined, and the definition parameters and the license plate definition are in negative correlation. The technical problem that the optimal license plate image cannot be accurately selected due to different definitions in the existing multi-angle license plate image is solved, and the technical effects of realizing high-accuracy license plate selection by customizing definition parameters and improving the license plate recognition accuracy and the traffic management automation efficiency are achieved.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and specifically to a method for selecting license plates based on sharpness comparison. Background Technology

[0002] In applications such as intelligent transportation, vehicle management, and urban security, license plate recognition technology has become an important means of acquiring vehicle information. Existing license plate recognition systems typically rely on surveillance cameras to continuously capture video of passing vehicles, extract the license plate area image, and then recognize the license plate characters. However, in actual acquisition processes, due to factors such as high-speed vehicle travel, changing shooting angles, unstable lighting conditions, and camera shake, the system often obtains multiple frames of license plate images of the same vehicle at the same time, and these images exhibit significant differences in clarity.

[0003] Automatically selecting the clearest frame from multiple images is crucial for improving license plate recognition accuracy. Existing methods typically employ traditional sharpness evaluation metrics such as edge strength, gradient changes, and frequency domain analysis. However, these methods are susceptible to noise interference in complex scenes, and some computations are complex and lack real-time performance. Furthermore, license plate images acquired from different angles and under different lighting conditions exhibit significant changes in content distribution, grayscale differences, and texture features, making it difficult for traditional sharpness algorithms to make stable and consistent sharpness judgments across multi-angle license plate images. Summary of the Invention

[0004] This application provides a license plate selection method based on clarity comparison, which solves the technical problem that existing multi-angle license plate images have varying clarity and cannot accurately select the best license plate image.

[0005] This application provides a license plate selection method based on clarity comparison, the method comprising:

[0006] A set of license plate images is obtained, wherein the set of license plate images is a group of multi-angle license plate images continuously captured from the same vehicle in the same scene; the sharpness parameters of each license plate image are calculated based on pixel vector operations to determine the sharpness parameter set; the minimum value of the sharpness parameter set is taken to determine the license plate selection result, wherein the sharpness parameter is negatively correlated with the sharpness of the license plate.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] First, multiple license plate images of the same vehicle taken consecutively from different angles in the same scene are acquired, forming a license plate image set. Then, for each license plate image in the set, its corresponding sharpness parameter is calculated using pixel vector calculation, and all sharpness parameters are combined into a parameter set. Since the calculated sharpness parameter is negatively correlated with the actual sharpness, the parameter with the smallest value is selected from the parameter set, and the corresponding license plate image is determined as the final sharpest license plate selection result for subsequent license plate recognition. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic flowchart of a license plate selection method based on clarity comparison, provided as an embodiment of this application.

[0011] Figure 2 This is a schematic diagram illustrating the calculation of clarity parameters in a license plate selection method based on clarity comparison, provided in an embodiment of this application. Detailed Implementation

[0012] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structure, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0013] Examples, such as Figure 1 As shown, this application provides a license plate selection method based on clarity comparison, the method including:

[0014] Acquire a set of license plate images, wherein the set of license plate images is a group of multi-angle license plate images continuously collected from the same vehicle in the same scene.

[0015] In this embodiment, continuous video frames are first captured from vehicles detected by monitoring equipment. After a vehicle enters the monitoring area, the camera acquires multiple frames containing the vehicle within a short period. Subsequently, the system extracts the corresponding license plate region from each frame of the vehicle image based on a license plate detection algorithm, and aggregates these continuously generated license plate region images from the same vehicle and the same acquisition scene to form a license plate image set. Since the vehicle's driving angle, license plate orientation, and lighting conditions constantly change during movement, the resulting license plate image set typically contains multiple license plate images with different shooting angles and different levels of clarity. These images collectively form the input basis for subsequent clarity evaluation and license plate selection.

[0016] Furthermore, by using monitoring equipment to continuously capture video frames of passing vehicles, a vehicle image set is determined; and by using a license plate monitoring algorithm, the license plate region of the vehicle image set is segmented to determine the license plate image set.

[0017] Preferably, the system first utilizes surveillance equipment deployed in scenarios such as roads, entrances / exits, or parking lots to continuously capture video frames of passing vehicles. These surveillance devices are typically high-definition cameras with automatic exposure, autofocus, and high-speed shutter capabilities. After a vehicle enters the monitoring field of view, the camera captures real-time images of the vehicle at a preset frame rate, such as 25fps or 30fps, recording multiple consecutively generated images. Since the vehicle is in motion, these consecutive frame images can cover the visual information of the vehicle at different positions and postures, thus forming a vehicle image set. After obtaining the vehicle image set, the system calls an existing license plate detection algorithm to automatically detect the license plate region in each frame of the vehicle image. This license plate detection algorithm can employ a deep learning detection model, such as YOLO, SSD, or other license plate localization networks based on convolutional neural networks, or it can use traditional image processing methods to identify color features, shape features, or texture features. The license plate detection algorithm first preprocesses the vehicle images, including scaling, normalization, or histogram equalization, and then determines the location range of the license plate in the image through feature extraction and region prediction steps. For the detected license plate area, the system further performs a region cropping operation, separating the rectangular region containing the license plate from the vehicle image to generate the corresponding license plate image. Then, the successfully segmented license plate areas from all consecutive frames are collected to form a set of license plate images of the same vehicle from multiple angles and with different resolutions, providing a reliable image data foundation for subsequent resolution comparison and optimal license plate selection.

[0018] The sharpness parameters are calculated for each license plate image in the set based on pixel vector operations to determine the sharpness parameter set.

[0019] In one embodiment, after obtaining the license plate image set, the images are first read pixel by pixel, converting all pixel values ​​in the image into a vector data sequence based on their grayscale intensity. Then, a statistical analysis of the overall brightness distribution of the image is performed based on this pixel vector, including calculating the mean grayscale value of all pixels and the standard deviation of the grayscale values ​​relative to the mean. Next, based on the calculated mean and standard deviation, a sharpness parameter corresponding to the current license plate image is generated through mathematical operations. This sharpness parameter comprehensively reflects the differences in pixel distribution and grayscale contrast characteristics of the image. By recording the sharpness parameter of each image sequentially, a sharpness parameter set can be compiled, providing a quantitative basis for subsequently selecting the license plate image with the optimal sharpness.

[0020] Furthermore, such as Figure 2 As shown, the calculation of sharpness parameters based on pixel vector operations includes:

[0021] For the first license plate image, calculate the pixel mean and pixel standard deviation, where the first license plate image is any license plate image in the set of license plate images; using the pixel mean as a boundary, calculate the first mean and the second mean and find the difference to determine the average difference, where the first mean is the average of pixels with pixel values ​​higher than the average, and the second mean is the average of pixels with pixel values ​​lower than the average; perform a division operation on the pixel standard deviation and the average difference to determine the first sharpness parameter.

[0022] Preferably, firstly, any license plate image is selected from the set of license plate images as the first license plate image to be processed. For this first license plate image, the grayscale values ​​of all its pixels are read, and the grayscale values ​​of each pixel are used as basic data in the calculation. Subsequently, the arithmetic mean of all pixel grayscale values ​​is processed to obtain the pixel mean, and the pixel standard deviation is calculated based on the deviation between all grayscale values ​​and the pixel mean, which is used to characterize the dispersion of the image brightness distribution. Then, using the calculated pixel mean as a dividing line, all pixels in the first license plate image are divided into two categories: one category is pixels with grayscale values ​​higher than the pixel mean, and the other category is pixels with grayscale values ​​lower than the pixel mean. Then, the arithmetic mean of the grayscale values ​​of the two categories of pixels is calculated separately to obtain the corresponding first mean and second mean, where the first mean reflects the average brightness level of pixels in bright areas; the second mean reflects the average brightness level of pixels in dark areas. By subtracting the first mean and the second mean, the average difference can be obtained, which can measure the degree of grayscale difference between bright and dark areas of the image. Finally, a parameter merging operation is performed, that is, the pixel standard deviation is used as the dividend and the average difference is used as the divisor to perform a division operation to obtain the first sharpness parameter of the current first license plate image. This first sharpness parameter combines the pixel distribution dispersion and the contrast characteristics of bright and dark areas. Its value is negatively correlated with the actual image sharpness, providing a quantitative basis for subsequent comparison of the sharpness between license plate images.

[0023] Furthermore, before calculating the sharpness parameters of each license plate image set based on pixel vector operations, the license plate image set is subjected to grayscale processing and size normalization processing.

[0024] Preferably, before calculating the sharpness parameters of the license plate image set, all license plate images in the set are preprocessed to ensure the accuracy and comparability of subsequent pixel vector calculations. Specifically, each license plate image is first converted to grayscale, transforming the original color license plate image into a single-channel grayscale image. During grayscale conversion, the red, green, and blue channel values ​​of each pixel in the image are weighted according to a preset ratio to obtain the corresponding grayscale value, thereby simplifying color information into brightness information, making subsequent pixel statistics and brightness comparison calculations more stable and consistent. After grayscale conversion, the license plate images are normalized. Since continuously acquired license plate images may have different resolutions or sizes due to changes in shooting distance, license plate orientation, and camera focal length, failure to process them will directly affect the comparability of the sharpness parameter calculations. Therefore, a uniform target size, such as a fixed width and height, is used to scale all license plate images, ensuring that all images maintain consistency in pixel count and spatial scale. During scaling, nearest neighbor interpolation, bilinear interpolation, or other image interpolation algorithms can be used to preserve the image content structure as much as possible. Through the above grayscale and size normalization operations, all images in the license plate image set have a unified grayscale space and size benchmark, which enables subsequent pixel vector-based sharpness parameter calculations to be performed under consistent conditions, ensuring the comparability and reliability of sharpness parameters between different license plate images.

[0025] Furthermore, in the calculation of the sharpness parameter, when the absolute value of the average difference is less than a preset threshold, a preset maximum value is used to replace the sharpness parameter.

[0026] Optionally, when calculating the sharpness parameter for each license plate image, a division operation is required using the difference in average values ​​and the standard deviation of pixels. However, in certain extreme cases, such as when the image is severely out of focus, severely overexposed or underexposed, or when noise interference causes the contrast between bright and dark areas to almost disappear, the pixel brightness distribution in the image will be highly concentrated, causing the average difference between bright and dark areas to approach zero. In such cases, directly performing a division operation may result in a division-to-zero error or an abnormally magnified, meaningless calculation result that cannot reflect the true sharpness of the image. To avoid these anomalies, a threshold protection mechanism is introduced when calculating the sharpness parameter. That is, when the absolute value of the average difference is detected to be lower than a preset threshold, such as a minimum value close to zero, it is determined that the license plate image can no longer provide effective brightness and contrast information and is an invalid image that is difficult to evaluate sharpness through pixel vector operations. At this time, the standard division result is no longer used, but the sharpness parameter is directly set to the preset maximum value. This preset maximum value is used to mark the image with extremely poor sharpness, so that it is always at the bottom in the subsequent sharpness ranking process, thereby automatically excluding extremely blurry, severely exposed, or contrast-depleted images, ensuring the reliability and stability of the final sharpness selection result.

[0027] The minimum value of the aforementioned clarity parameters is taken to determine the license plate selection result, wherein the clarity parameters are negatively correlated with the clarity of the license plate.

[0028] In one embodiment, after calculating the sharpness parameters of all images in the license plate image set, these sharpness parameters are summarized to form a sharpness parameter set. Since the constructed sharpness parameters are negatively correlated with the actual image sharpness—that is, the sharper the image, the smaller the calculated sharpness parameter value; conversely, the blurrier the image, the larger the parameter value—the sharpness parameter set is iterated and compared to select the sharpness image with the lowest value from the license plate image set. Then, based on the index position of this lowest sharpness parameter in the parameter set, the corresponding license plate image is located, and this image is determined as the selected license plate. In this way, without manual intervention, the most suitable high-quality license plate image for subsequent character recognition can be accurately identified from candidate license plate images of multiple angles and sharpness levels, thereby improving the overall performance and stability of license plate recognition.

[0029] Furthermore, the license plate selection result is determined by taking the minimum value from the set of clarity parameters, including:

[0030] The set of sharpness parameters is sorted from largest to smallest to determine the sharpness parameter sequence; in the sharpness parameter sequence, the smallest sharpness parameter at the end of the sequence is taken, and the corresponding license plate image is used as the license plate selection result.

[0031] Preferably, to facilitate accurate selection of the license plate image with the highest clarity from the set of clarity parameters, the aforementioned set of clarity parameters is first sorted. Specifically, all clarity parameters in the set are sorted from largest to smallest, generating a clarity parameter sequence. Since the clarity parameters are negatively correlated with the clarity of the actual image, the parameters at the beginning of the sequence correspond to license plate images with lower clarity, while the parameters at the end of the sequence correspond to license plate images with the best clarity. After obtaining the sorted clarity parameter sequence, the clarity parameter at the end of the sequence is read and used as the minimum clarity parameter. Then, based on the corresponding position of this minimum clarity parameter in the original parameter set before sorting, the specific license plate image that matches it is located. This license plate image is the optimal license plate image automatically selected, i.e., the license plate selection result. Through this sorting and selection process, the target image with the highest clarity can be found efficiently and reliably among multiple frames, multiple angles, and license plate images with varying clarity, providing the highest quality input image for subsequent license plate character recognition.

[0032] Furthermore, after determining the license plate selection result, the following steps are taken:

[0033] Read the license plate selection result to determine the candidate license plate images; input the candidate license plate images into the license plate recognition system for license plate character recognition.

[0034] Preferably, after determining the license plate selection result based on the sharpness parameter sequence, this result is read and stored or transmitted to the license plate recognition system as a candidate license plate image. Since the candidate license plate image is the image with the highest sharpness automatically selected through a sharpness comparison process, it outperforms other candidate images in terms of visual quality, character edge sharpness, and texture detail, thus significantly improving the stability and accuracy of subsequent character recognition. Subsequently, the license plate recognition system can employ traditional character segmentation and template matching methods, or it can use an end-to-end license plate recognition model based on deep learning to perform character feature extraction and classification recognition on the license plate image, ultimately outputting a complete license plate number string. Using the license plate image with the highest sharpness effectively reduces recognition errors caused by blurring, ghosting, or insufficient contrast during character recognition, thereby significantly improving the performance of the entire license plate recognition process.

[0035] Furthermore, a correlation is established between the evaluation results of the clarity parameter and the acquisition parameters of the monitoring equipment. Based on the correlation, when the clarity parameter fails to meet the standard in the license plate selection results determined from N consecutive sets of license plate images, an acquisition control command is automatically sent to the monitoring equipment to adjust the acquisition parameters of the monitoring equipment.

[0036] Preferably, during long-term operation, the calculation results of the license plate image sharpness parameters are continuously recorded and stored in association with the acquisition parameters of the monitoring equipment at the time the result is generated. These acquisition parameters include, but are not limited to, exposure time, gain value, shutter speed, focal length, aperture size, and image enhancement mode. By analyzing the correspondence between the sharpness parameters and the imaging effects under different acquisition parameters, a dynamic model or empirical rule base reflecting the relationship between the equipment's acquisition quality and image sharpness can be constructed; that is, a correlation between the sharpness parameter evaluation results and the acquisition parameters is established. After the correlation is established, the sharpness of multiple sets of license plate images continuously acquired by the monitoring equipment is monitored in real time. When N consecutive sets of license plate images (N can be preset or dynamically adjusted according to the actual scene) are compared for clarity, and the clarity parameters of the final license plate selection result do not meet the preset clarity qualification threshold, it is determined that there may be a deviation in the current acquisition status of the monitoring equipment, such as underexposure, excessive brightness, inaccurate focus, or sudden changes in ambient light leading to a decrease in image quality. At this time, based on the previously established correlation between clarity and acquisition parameters, the acquisition factors that may lead to a decrease in clarity are identified, and corresponding acquisition control instructions are generated. For example, when insufficient clarity is detected and the exposure value deviates from the exposure parameters of the historical best image, an exposure adjustment instruction can be automatically generated; if a focus shift is detected, an autofocus or focus correction instruction can be automatically issued; if the ambient light is too dark, the gain can be increased or the supplementary lighting function can be activated. The generated control instructions will be sent to the monitoring equipment in real time, and the monitoring equipment will automatically execute parameter adjustment operations to optimize the subsequent image acquisition quality. Through the above mechanism, a closed-loop linkage between clarity evaluation and acquisition control is realized, enabling the monitoring equipment to automatically correct the acquisition parameters when the imaging quality deteriorates, thereby continuously maintaining the high-quality imaging capability of license plate acquisition images and improving the stability and reliability of the license plate recognition system in complex environments.

[0037] In summary, the embodiments of this application have at least the following technical effects:

[0038] First, a set of license plate images is acquired, which consists of a group of multi-angle license plate images continuously captured from the same vehicle in the same scene. Next, a sharpness parameter is calculated for each license plate image in the set based on pixel vector operations to determine a sharpness parameter set. Finally, the minimum value in the sharpness parameter set is taken to determine the license plate selection result, where the sharpness parameter is negatively correlated with the license plate sharpness. This solves the technical problem of inconsistent sharpness in existing multi-angle license plate images, making it impossible to accurately select the best license plate image. It achieves high-accuracy license plate selection through customized sharpness parameters, improving license plate recognition accuracy and traffic management automation efficiency.

[0039] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A license plate selection method based on clarity comparison, characterized in that, The method includes: Acquire a set of license plate images, wherein the set of license plate images is a group of multi-angle license plate images continuously collected from the same vehicle in the same scene; The sharpness parameters of each license plate image are calculated based on pixel vector operations to determine the sharpness parameter set. The minimum value of the aforementioned clarity parameters is taken to determine the license plate selection result, wherein the clarity parameters are negatively correlated with the clarity of the license plate.

2. The license plate selection method based on sharpness comparison as described in claim 1, wherein the sharpness parameter calculation based on pixel vector operation includes: For the first license plate image, calculate the pixel mean and pixel standard deviation, wherein the first license plate image is any license plate image in the set of license plate images; Using the pixel mean as a boundary, calculate the first mean and the second mean and find the difference to determine the average difference. The first mean is the average of pixels with pixel values ​​higher than the average, and the second mean is the average of pixels with pixel values ​​lower than the average. The first sharpness parameter is determined by dividing the difference between the pixel standard deviation and the average value.

3. The license plate selection method based on clarity comparison as described in claim 1, characterized in that, Before performing pixel vector-based resolution parameter calculations on each of the license plate image sets, the license plate image sets are subjected to grayscale conversion and size normalization.

4. The license plate selection method based on clarity comparison as described in claim 1, characterized in that, By using monitoring equipment, continuous video frames are captured of passing vehicles to determine the vehicle image set; Based on the license plate detection algorithm, the license plate region is segmented in the vehicle image set to determine the license plate image set.

5. The license plate selection method based on clarity comparison as described in claim 2, characterized in that, In the calculation of the sharpness parameter, when the absolute value of the average difference is less than a preset threshold, a preset maximum value is used to replace the sharpness parameter.

6. The license plate selection method based on clarity comparison as described in claim 1, characterized in that, The minimum value of the aforementioned clarity parameters is taken to determine the license plate selection result, including: The sharpness parameter set is sorted from largest to smallest to determine the sharpness parameter sequence; In the resolution parameter sequence, the smallest resolution parameter at the end of the sequence is taken, and the corresponding license plate image is used as the license plate selection result.

7. The license plate selection method based on clarity comparison as described in claim 1, characterized in that, After the license plate selection results are confirmed, the following will be included: Read the license plate selection results and determine the candidate license plate images; The candidate license plate images are input into the license plate recognition system for license plate character recognition.

8. The method as described in claim 1, characterized in that, Establish the correlation between the evaluation results of the resolution parameters and the acquisition parameters of the monitoring equipment; Based on the aforementioned correlation, when the clarity parameter fails to meet the standard in the license plate selection results determined from N consecutive sets of license plate images, an acquisition and control command is automatically sent to the monitoring equipment to adjust the acquisition parameters of the monitoring equipment.

Images