Highway traffic image enhancement method under complex weather

Abstract

The present application relates to a kind of expressway traffic image enhancement method under complex weather, compared with prior art, solve the defects of poor image recovery quality and unstable performance.The present application includes the following steps: obtaining the degraded image of expressway under severe weather;Calculate the attenuation coefficient of degraded image;Based on the attenuation coefficient, calculate the detail layer of degraded image;Enhance the detail layer of degraded image;Traffic image enhancement image is obtained.The present application is based on adaptive Gaussian filtering and Sigmoid transformation, which can effectively improve the appearance color of image and enhance the contrast, brightness and clarity of image, providing more reliable visual information for realizing all-weather, full-network traffic situation, traffic anomaly, road construction and other state precision perception, through the high-quality image after enhancement, improving the precision of traffic target detection and traffic state discrimination, improving the accuracy of traffic accident prediction under severe weather.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traffic image processing, in particular to a highway traffic image enhancement method under complex weather. BACKGROUND

[0002] In the field of intelligent driving, traffic situation awareness, etc., with the increase of traffic flow, accidents on the highway are becoming more and more common, and the impact is also getting bigger and bigger. As an important information carrier, images often appear color degradation, contrast reduction, and detail blur due to weather influences such as fog, haze, and sandstorms, resulting in reduced target recognition accuracy. It is essential to provide high-quality input features for traffic state monitoring systems and traffic accident prediction models by proposing a simple and effective image enhancement method.

[0003] Overall, the methods for enhancing degraded images can be divided into three categories: image enhancement methods based on physical models, image enhancement methods based on non-physical models, and image enhancement methods based on deep learning. Among them, the image enhancement method based on physical model first establishes a mathematical model according to the imaging mechanism of the degraded image in complex weather, and analyzes the unknown parameters in the model according to the hand prior knowledge or hypothesis condition, and inversely deduces the enhanced clear image. This method can achieve certain enhancement effect in a specific degraded scene, but due to the harshness and variability of complex weather, the stability of prior knowledge is unstable, and the robustness and stability of the enhancement effect are not the same.

[0004] The image enhancement method based on non-physical model often modifies the pixel value or histogram distribution of the image to improve the overall or local contrast information of the image, improve the color of the image, and achieve image enhancement. Although this method has low computational complexity, it lacks consideration of the degradation mechanism of the degraded image, resulting in insufficient details in the enhanced result.

[0005] The image enhancement method based on deep learning uses the powerful feature learning and operation capability of deep learning network to obtain the final enhanced image. This method can improve the visual quality of the degraded image to a certain extent, but it usually needs to rely on a large number of paired "degraded image-clear image" data sets to train the deep learning network, and these data sets are difficult to obtain under complex and harsh weather conditions, causing the limitation of this method to be larger.

[0006] In particular, existing image enhancement techniques, such as atmospheric scattering correction based on physical models, histogram equalization based on non-physical models, or end-to-end restoration based on deep learning, have certain effects in general scenarios, but are difficult to adapt to the unique degradation characteristics of highway traffic images.

[0007] The key targets such as license plates and road signs have a very small proportion in images, and the traditional global enhancement method (such as histogram stretching) will excessively amplify the background noise, resulting in the target being submerged; the same picture has vehicle details and heavy fog, and the fog concentration and snow intensity change dramatically with the spatial position, and the fixed parameter physical model cannot accurately model the local attenuation traffic signal lights, warning signs and other color information directly related to machine decision-making, and the existing enhancement method is easy to introduce color deviation or saturation distortion when improving the contrast, such as white red signal lights, causing machine misjudgment.

[0008] The above problems result in the phenomenon that the restoration results of the prior art generally have low light intensity, unclear lane lines and unnatural colors in the highway scene, and cannot realize accurate perception of traffic situation, traffic anomaly, road construction and other states in all-weather and all-road networks, and it is difficult to meet the stringent requirements of intelligent transportation systems on image reliability. Therefore, how to design a method for robustly restoring traffic image colors and details under complex weather conditions has become a technical problem that needs to be solved urgently. SUMMARY

[0009] The purpose of the present application is to solve the defects of poor image restoration quality and unstable performance in the prior art, and to provide a highway traffic image enhancement method under complex weather conditions to solve the above problems.

[0010] In order to achieve the above purpose, the technical scheme of the present application is as follows:

[0011] A highway traffic image enhancement method under complex weather conditions, comprising the following steps:

[0012] 11) Obtain a degraded image under severe weather on a highway;

[0013] 12) Calculate the attenuation coefficient of the degraded image;

[0014] 13) Calculate the detail layer of the degraded image based on the attenuation coefficient;

[0015] 14) Enhance the detail layer of the degraded image;

[0016] 15) Obtain a traffic image enhancement image: restore the natural color of the enhanced detail layer of the degraded image.

[0017] The calculation of the attenuation coefficient of the degraded image comprises the following steps:

[0018] 21) Divide the L channel of the degraded image into local pixel blocks, calculate the mean value and variance of each local pixel block, and the specific formula is:

[0019] ,

[0020] ,

[0021] where, represents the pixel value of the local pixel block of the degraded image L channel at position coordinates , represents the top-left corner coordinates of the image block, and M and N represent the width and height of the local pixel block, respectively;

[0022] 22) Smooth the local mean and global variance over the whole image by guided filtering: :

[0023] ,

[0024] ,

[0025] where, represents and the position coordinates over the whole image, represents the guided filtering, represents the luminance component of the traffic image in the LAB color space;

[0026] 23) Calculate the attenuation distribution : using the mean and variance at

[0027] :

[0028] where max() and min() are the maximum and minimum functions, respectively;

[0029] 24) Calculate the auxiliary matrix and based on the attenuation distribution:

[0030] ,

[0031] ,

[0032] where is the maximum value of the attenuation distribution ;

[0033] Calculate the matrix :

[0034] ,

[0035] Calculate the attenuation coefficient:

[0036] ​ ,

[0037] where, is the attenuation coefficient, is the matrix of the mean value, H and W are the width and height of the whole image respectively.

[0038] The step of calculating the detail layer of the degraded image based on the attenuation coefficient comprises the following steps:

[0039] 31) The base layer of the degraded image is obtained by convolving the high-pass filter with

[0040] ,

[0041] where, represents and the position coordinates on the whole image, represents the base layer of the degraded image, is the attenuation coefficient is the high-pass filter, is the local mean value, is the global variance, is the degraded image with position coordinates represented;

[0042] 32) The detail layer is obtained by using the difference between the degraded image and the base layer:

[0043] ,

[0044] where, is the detail layer of the degraded image, is the weight coefficient:

[0045] ,

[0046] where, is the attenuation distribution.

[0047] The step of enhancing the detail layer of the degraded image comprises the following steps:

[0048] 41) The local contrast and Laplacian variance of the degraded image are calculated:

[0049] ,

[0050] ,

[0051] where, is the local contrast of the degraded image, is the Laplacian variance of the degraded image, ​is the result of the convolution of the degraded image with the Laplacian filter, is the mean value of is the degraded image with position coordinate representation, is the local mean value over the whole image, H and W are the width and height of the whole image respectively, represents and the position coordinate over the whole image;

[0052] 42) Calculate the enhancement coefficient of the detail layer of the degraded image :

[0053] ;

[0054] 43) Enhance the detail layer of the degraded image based on the enhancement coefficient :

[0055] ,

[0056] wherein, is the enhanced detail layer of the degraded image with position coordinate representation, is the detail layer of the degraded image.

[0057] The obtaining of the traffic image enhanced image comprises the following steps:

[0058] 51) Design the local sensitivity coefficient of human eye visual perception :

[0059] ,

[0060] wherein, is the decay distribution, represents and the position coordinate over the whole image;

[0061] 52) Design the global sensitivity coefficient of human eye visual perception :

[0062] ,

[0063] wherein, is the threshold value, is the enhanced detail layer of the degraded image;

[0064] 53) Synthesize the local sensitivity coefficient and the global sensitivity coefficient to obtain the comprehensive sensitivity coefficient of human eye visual perception to color and contrast : ​

[0065] ;

[0066] 54)Application of Sigmoid transformation to obtain the final enhanced image :

[0067] ,

[0068] wherein, refers to an exponential with base and variable .

[0069] A computer readable storage medium, the storage medium has a computer program stored thereon, when the computer program is executed by a processor, a complex weather highway traffic image enhancement method can be realized.

[0070] A computer device, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, when the processor executes, a complex weather highway traffic image enhancement method can be realized.

[0071] Advantages

[0072] The complex weather highway traffic image enhancement method of the application, compared with the prior art, based on adaptive Gaussian filtering and Sigmoid transformation, can effectively improve the appearance color of the image and enhance the contrast, brightness and definition of the image, provide more reliable visual information for realizing all-weather, all-road network traffic situation, traffic anomaly, road construction and other state precise perception, improve the precision of traffic target detection and traffic passing state discrimination through the enhanced high-quality image, and improve the accuracy of traffic accident prediction in bad weather.

[0073] The application utilizes local mean and local variance to design an approximate representation factor of image attenuation degree, and establishes an adaptive Gaussian filter to eliminate redundant color cast information and extract more interesting structure and texture image; then the detail layer of the image is converted to gradient domain expression, and the global mean of the attenuation factor is used to stretch the gradient information of different color channels, so as to capture the significant details and contour information in the image; finally, according to the information perception mechanism of human eye retina, a Sigmoid mapping function is designed to dynamically simulate the nonlinear mapping relationship of human color perception, so as to realize the correction of dynamic range and color of the enhanced image. BRIEF DESCRIPTION OF DRAWINGS

[0074] Figure 1 is the sequence diagram of the method of the application;

[0075] Figure 2 is the algorithm flowchart of the application;

[0076] Figure 3 This is a comparison diagram of the effects involved in the present invention. Detailed Implementation

[0077] To provide a better understanding of the structural features and effects achieved by the present invention, a detailed description is provided below, accompanied by preferred embodiments and accompanying drawings:

[0078] like Figure 1 and Figure 2 As shown, the present invention provides a method for enhancing highway traffic images under complex weather conditions, comprising the following steps:

[0079] The first step is to acquire degraded images of highways under severe weather conditions.

[0080] The second step is to calculate the attenuation coefficient of the degraded image.

[0081] Due to the complexity and high scene dependence of traffic scene image degradation patterns (e.g., significant differences in light attenuation, scattering, and noise characteristics caused by different environmental factors such as fog, dust, and low light), a single fixed attenuation coefficient model is difficult to effectively adapt to all degradation situations. Designing specific attenuation coefficients for different scenes aims to estimate the degree of degradation in each scene. This targeted design can more accurately estimate the scene-related degree of degradation, thereby achieving more refined compensation in the subsequent restoration process. This effectively avoids overcompensation or undercompensation problems caused by a single model, ultimately significantly improving the quality, realism, and usability of the restored image in subsequent intelligent transportation system applications (such as license plate recognition and behavior analysis).

[0082] (1) Degraded image The L channel is divided into local pixel blocks, and the mean value of each local pixel block is calculated. and variance The specific formula is as follows:

[0083] ,

[0084] ,

[0085] in, Represents the position coordinates of local pixel blocks in the L channel of the degraded image. Pixel value at that location, This represents the coordinates of the top-left vertex of the image block, where M and N represent the width and height of the local pixel block, respectively.

[0086] (2) Smooth the local mean of the entire image by guided filtering. and global variance :

[0087] ,

[0088] ,

[0089] wherein, denotes and the position coordinates on the whole image, denotes the guided filtering, denotes the brightness component of the traffic image in the LAB color space.

[0090] (3) the attenuation distribution is calculated by using the mean and the variance at the position :

[0091] ,

[0092] wherein, max() and min() are the maximum and minimum functions respectively.

[0093] (4) the auxiliary matrix and are calculated based on the attenuation distribution:

[0094] ,

[0095] ,

[0096] wherein is the maximum value of the attenuation distribution ;

[0097] the matrix is calculated:

[0098] ,

[0099] the attenuation coefficient is calculated:

[0100] ,

[0101] wherein, is the attenuation coefficient, is the mean value of the matrix , and H and W are the width and height of the whole image respectively.

[0102] Step 3, the detail layer of the degradation image is calculated based on the attenuation coefficient.

[0103] ​In traffic images under complex weather conditions, key detail information (such as license plate characters, traffic sign edges, pedestrian contours, and vehicle textures) is usually mixed with noise, artifacts, and overall illumination / color degradation in the same or similar frequency range. The core purpose of separating the detail layer through a specially designed step is to physically isolate and highlight the high-frequency information component of the image. This separation enables subsequent restoration algorithms to accurately and specifically enhance and restore detail information representing the structure of the real scene (edges, textures), while minimizing the risk of misoperation on noise or low-frequency degradation components (such as large-area color bias and uniform blur) mixed therein.

[0104] (1) The base layer of the degraded image is obtained by convolution of the Gaussian filter and

[0105]

[0106] wherein, represents and the position coordinates of the image, represents the base layer of the degraded image, is a Gaussian filter with a decay coefficient of is the local mean, is the global variance, is the degraded image with position coordinates.

[0107] (2) The detail layer is obtained by using the difference between the degraded image and the base layer:

[0108]

[0109] wherein, is the detail layer of the degraded image, is the weight coefficient:

[0110]

[0111] wherein, is the decay distribution.

[0112] Fourth step, enhance the detail layer of the degraded image.

[0113] By enhancing the detail layer of the degraded image, the controllability and effectiveness of the restoration process are significantly improved, and the final restored image is qualitatively improved in terms of clarity, detail richness, and visual usability, which is particularly beneficial to the accuracy improvement of subsequent intelligent traffic systems (such as license plate recognition, violation detection, and behavior analysis).

[0114] (1) Calculate the local contrast and Laplacian variance of the degraded image:​​​​​

[0115] ,

[0116] ,

[0117] wherein, is the local contrast of the degraded image, is the Laplacian variance of the degraded image, is the result of the degraded image convolved with the Laplacian filter, is the mean value of , is the degraded image with position coordinate representation, is the local mean value over the whole image, H and W are the width and height of the whole image respectively, represents and the position coordinate over the whole image.

[0118] (2) Calculate the enhancement coefficient of the detail layer of the degraded image :

[0119] .

[0120] (3) Enhance the detail layer of the degraded image based on the enhancement coefficient :

[0121] ,

[0122] wherein, is the enhanced detail layer of the degraded image with position coordinate representation, is the detail layer of the degraded image.

[0123] Fifth step, obtaining the enhanced traffic image: restoring the natural color of the enhanced detail layer of the degraded image. In the restoration process of the degraded traffic image, although the enhancement operation on the detail layer can significantly improve the clarity, it is easy to introduce nonlinear chroma distortion, which is manifested as color deviation, unnatural color spots or abnormal saturation in the enhanced area. This distortion is particularly harmful in traffic scenes containing key color information (such as traffic signal light status, warning sign color, vehicle tail light, special painting), which may lead to misjudgment of the subsequent intelligent analysis system (such as violation judgment, event detection). Therefore, it is crucial to specially design a step to restore the natural color of the enhanced detail layer. The core purpose is to correct the local chroma distortion caused by the high-frequency detail enhancement algorithm, by constraining the color change within a reasonable range that conforms to the physical illumination and inherent properties of objects, to ensure that the restored image is not only clear and sharp, but also has high color fidelity and visual naturalness, thereby ensuring the reliability and decision accuracy of the restored result in traffic monitoring and analysis.

[0124] (1) Designing the local sensitivity coefficient of human eye visual perception :

[0125] ,

[0126] wherein, is the attenuation distribution, represents and the position coordinates on the whole image.

[0127] (2) Designing the global sensitivity coefficient of human eye visual perception :

[0128] ,

[0129] wherein, is the threshold value, is the detail layer of the enhanced degraded image.

[0130] (3) Integrating the local sensitivity coefficient and the global sensitivity coefficient to obtain the comprehensive sensitivity coefficient of human eye visual perception to color and contrast :

[0131] .

[0132] (4) Applying the Sigmoid transformation to obtain the final enhanced image :

[0133] ,

[0134] wherein, is an index with as the base and as the variable.

[0135] In order to verify the effectiveness of the highway traffic image enhancement method under complex weather conditions according to the present application, several common traffic degraded images are selected for testing and verifying the enhancement performance, namely (a), (b), (c) and (d) in Figure 3 . It can be seen that the original traffic images have complex degradation phenomena such as fogging, color cast and low illumination; the DCP method can eliminate the fogging effect of the light fog image, as shown in (c) in Figure 3 and (d) in Figure 3 , the contrast of the processed image is improved, but for (a) in Figure 3 and (b) in Figure 3 , the color cast and low illumination effects cannot be eliminated.In the low-light or foggy traffic image shown in (b), this method failed to achieve satisfactory enhancement and further aggravated the color cast. Similarly, the CAP method can only effectively process traffic images in lightly foggy scenes. Compared to the above two methods, the DAACC method performs better in eliminating fog effects and has a more significant contrast enhancement effect. However, because it does not consider the complexity and high scene dependence of the degradation patterns in traffic scene images under complex traffic scenes, the processing results show large areas of dark areas and a bluish-green background, such as... Figure 3 (c) and Figure 3 As shown in (d), the BCDP method has limited enhancement effects on these complex traffic images and introduces additional color cast phenomena, such as... Figure 3 The sky area shown in (b) has a reddish hue. Figure 3 (c) and Figure 3 In image (d), a yellowish background tone appears. MR methods can only reduce the hazy background, and the processed image still exhibits a certain degree of residual haze, resulting in poor detail display. Furthermore, these methods are ineffective in improving brightness and cannot meet the needs of traffic image enhancement under complex weather conditions such as overcast or foggy days. In contrast, the method of this invention fully considers the degradation mechanisms of traffic images under complex weather conditions, such as illumination, absorption, and scattering. It effectively extracts high-frequency detail information of key traffic elements (such as lane lines and vehicles) by precisely stripping the detail layer, while filtering out low-frequency noise and color casts from the original degraded image. Finally, by utilizing the visual mechanisms of human vision, the processing result of this invention has natural and realistic color saturation, rich detail, and sufficient brightness. Key road information such as vehicles, lane lines, and LCD signage are clearly visible, ensuring the integrity and reliability of the layered processing mechanism of this invention in enhancing the quality of degraded traffic images, thus meeting the analysis needs of intelligent transportation systems.

[0136] Furthermore, this invention utilizes three evaluation metrics to comprehensively evaluate the image enhancement performance of different methods: a block-based contrast quality index (PCQI), and an index for adding visible edges (…). ) and the normalized gradient mean of visible edges ( The three evaluation metrics for the enhancement performance of each method on the test images are shown in Tables 1, 2, and 3. It can be seen that the method of this invention achieves optimal performance on all evaluation metrics on the test images, significantly outperforming current mainstream image enhancement methods. Specifically, compared with the suboptimal method, the PCQI metric of this invention is improved by at least 3.09% and at most 34.75%, fully demonstrating the effectiveness and robustness of this invention in enhancing the contrast quality of traffic images under complex weather conditions; simultaneously, this invention… and The index is respectively increased by at least 14.32% and 80.38%, and is increased by at most 64.93% and 314.68%, which indicates that the method has excellent performance in enhancing the key details of the traffic image and the color performance of the image.

[0137] Table 1 PCQI index comparison of enhancement performance of each method on the test image

[0138] DCP CAP DAACC BCDP MR The present invention (a) 1.139 1.039 1.000 0.987 1.002 1.477 (b) 0.779 0.801 0.876 0.892 0.761 1.202 (c) 0.880 0.748 0.590 1.053 0.780 1.097 (d) 0.953 0.869 0.761 1.067 0.911 1.100

[0139] Table 2 PCQI index comparison of enhancement performance of each method on the test image Index quantitative comparison

[0140] DCP CAP DAACC BCDP MR The present invention (a) 4.865 1.904 4.377 2.449 0.980 5.948 (b) 2.049 0.519 0.968 3.117 0.190 5.141 (c) 0.211 0.337 0.095 0.167 0.218 0.400 (d) 0.284 0.281 0.257 0.349 0.243 0.399

[0141] Table 3 PCQI index comparison of enhancement performance of each method on the test image Index quantitative comparison

[0142] DCP CAP DAACC BCDP MR The present invention (a) 3.056 1.667 2.790 1.689 1.477 11.181 (b) 0.989 0.874 2.124 1.559 0.713 8.808 (c) 1.101 1.002 1.314 1.321 0.974 2.618 (d) 1.291 1.201 1.431 1.489 1.170 2.686

[0143] The basic principle, main features and advantages of the present application are shown and described above. It should be understood by those skilled in the art that the present application is not limited by the above examples, and the above examples and descriptions in the specification are only the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection required by the present application is defined by the appended claims and their equivalents.

Claims

1. A method for enhancing highway traffic images under complex weather conditions, characterized in that, The method comprises the following steps: 11) obtaining a degraded image under bad weather on the highway; 12) calculating the attenuation coefficient of the degraded image; The calculation of the attenuation coefficient of the degraded image comprises the following steps: 121) dividing the L channel of the degraded image into local pixel blocks, calculating the mean and variance of each local pixel block, in particular according to the formulas: ， ， wherein, represents a pixel value of a local pixel block of a degraded image L channel at a position coordinate , represents a top-left corner vertex coordinate of the image block, and M and N represent a width and a height of the local pixel block, respectively; 122) Smoothing the local mean over the whole image by guided filtering and global variance : ， ， wherein denotes and position coordinates over the entire image, denotes guided filtering, denotes the luminance component of the traffic image in the LAB color space; 123) using the mean and variance at the location to calculate the attenuation profile : ， Wherein, max() and min() are the maximum and minimum functions, respectively; 124) computing an auxiliary matrix based on the decay profile and : ， ， wherein is the maximum value of the decay profile of the decay profile Computing matrix : ， The attenuation coefficient is calculated as follows: ， wherein, is the attenuation coefficient, is the matrix is the mean value of the matrix H and W are the width and height of the whole image, respectively; 13) calculating the detail layer of the degraded image based on the attenuation coefficient; The calculation of the detail layer of the degraded image based on the attenuation coefficient comprises the following steps: 131) by a Gaussian filter with convolution to obtain a base layer of the degraded image: ， in, express and Position coordinates on the entire image, The base layer representing the degraded image. The attenuation coefficient is Gaussian filter, It is a local mean. For global variance, A degraded image with positional coordinates; 132) obtaining the detail layer by using the difference between the degraded image and the base layer: ， wherein is a detail layer of the degraded image, is a weight coefficient: ， wherein is an exponential decay distribution; 14) enhancing the detail layer of the degraded image; 15) obtaining the enhanced image of the traffic image: restoring the natural color of the enhanced detail layer of the degraded image.

2. The method of claim 1, wherein the method is a method of enhancing a highway traffic image under complex weather, characterized by, The enhancement of the detail layer of the degraded image comprises the following steps: 21) calculating the local contrast and Laplacian variance of the degraded image: ， ， wherein, is the local contrast of the degraded image, is the Laplacian variance of the degraded image, is the result of the degraded image convolved with the Laplacian filter, is the mean of is the degraded image with position coordinate representation, is the local mean over the whole image, H and W are the width and height of the whole image, respectively, represents and is the position coordinate over the whole image.​ 22) calculating an enhancement coefficient for the degraded image detail layer : ； 23) based on an enhancement coefficient enhancing the detail layer of the degraded image: ， wherein, is the enhanced detail layer of the degraded image with position coordinate representation, is the detail layer of the degraded image.

3. The method of claim 1, wherein the method is a method of enhancing a highway traffic image under complex weather, characterized by, The obtaining of the enhanced image of the traffic image comprises the following steps: 31) Designing the local sensitivity coefficient of human eye visual perception : ， wherein is an attenuation distribution, denotes and position coordinates over the entire image; 32) Designing global sensitivity coefficients of human eye visual perception : ， wherein, is a threshold value, is a detail layer of the enhanced degraded image; 33) the integrated local sensitivity coefficient and the global sensitivity coefficient to obtain the integrated sensitivity coefficient of human eye vision to color and contrast perception : ； 34) Apply Sigmoid transform to get final enhanced image : ， wherein is an exponential with base and variable .

4. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, and when the computer program is executed by the processor, the method for enhancing the traffic image on the highway under complex weather according to any one of claims 1-3 can be realized.

5. A computer device, comprising: The computer program stored in the memory and executable on the processor, when the processor is executed, the method for enhancing the traffic image on the highway under complex weather according to any one of claims 1-3 can be realized.

Images