Intelligent traffic task execution method and device based on image defogging

Abstract

The invention provides an intelligent traffic task execution method and device based on image defogging. The method comprises the steps of collecting a road image and calculating atmospheric light of the road image. And screening out the minimum value of a plurality of color channels of each pixel point in the road image to obtain a minimum color channel. And calculating the haze concentration of the road image, introducing a correction coefficient, constructing a nonlinear correction function, and mapping the haze concentration by using the nonlinear correction function to obtain the transmissivity. And inputting the transmissivity and the atmospheric light into an atmospheric scattering model for inversion so as to remove haze in the road image and obtain a defogged image. And inputting the defogged image into an intelligent traffic system, and executing an intelligent traffic task. The defogging algorithm adopted by the invention does not depend on mass data, is efficient in calculation and is suitable for complex real scenes, and the intelligent traffic task execution method based on the defogging algorithm can accurately carry out license plate recognition and target tracking.

Description

Technical Field

[0001] The present invention relates to the technical field of image defogging, and in particular to an intelligent traffic task execution method and device based on image defogging. Background Art

[0002] Under severe weather conditions such as haze, the contrast of images is reduced and colors are distorted due to the scattering effect of suspended particles in the atmosphere, which seriously affects the effect of subsequent visual tasks. Existing dehazing algorithms are mainly divided into two categories: the first category is based on physical models, and the second category is based on deep learning. However, there are the following problems: (1) Methods based on physical models, such as dark channel prior and fog line prior, rely on strong assumptions and are prone to failure in complex scenes, resulting in over-dehazing or color distortion. Algorithms such as histogram equalization and Retinex enhancement can improve the contrast of haze images, but they are prone to over-enhancement of haze images or loss of details in haze images. (2) Methods based on deep learning require a large amount of labeled data, have poor generalization ability in real scenes, and consume a lot of hardware resources. Dehazing algorithms based on convolutional neural networks (CNN) or generative adversarial networks (GAN) can better restore image quality, but their interpretability is low and they require high computing resources, making it difficult to run in real time on low-power devices.

[0003] Image dehazing technology is widely used in intelligent transportation systems. In haze weather conditions, the suspended particles in the atmosphere are affected by scattering and attenuation during the imaging process, resulting in blurring, contrast reduction, color distortion and other degradation problems in the images and videos collected by traffic monitoring equipment, which seriously affects the performance of the intelligent transportation system. For example, when the intelligent transportation system is affected by haze, there are the following problems: (1) Failure of license plate recognition: haze reduces the image clarity, making it difficult for traditional character recognition technology to accurately interpret license plate characters, thereby reducing the reliability of applications such as violation capture, parking management and electronic toll collection systems. (2) Intelligent monitoring misjudgment: low-quality images collected by traffic cameras may increase the misjudgment rate of tasks such as pedestrian detection, vehicle classification, and traffic flow statistics, affecting the accuracy of intelligent signal-based control methods and scheduling decisions. (3) Interference in target trajectory tracking and accident detection: due to the fuzzy target contour, it is difficult for vision-based target detection and tracking algorithms to correctly identify and predict the trajectory of vehicles or pedestrians, which affects the accuracy and timeliness of accident warning, congestion analysis and emergency response.

[0004] Therefore, there is a need for a single-frame image defogging algorithm that does not rely on massive data, is computationally efficient, and is applicable to complex real-world scenes. In addition, a method for executing intelligent traffic tasks based on this defogging algorithm can accurately parse license plate characters, realize intelligent monitoring, and accurately track target trajectories. Summary of the invention

[0005] To overcome the problems existing in the related technologies, the purpose of the present invention is to provide an intelligent transportation task execution method and device based on image dehazing, wherein the method can accurately parse license plate characters, achieve intelligent monitoring, and accurately track the target trajectory, and the single-frame image dehazing algorithm used in the intelligent transportation task execution method does not rely on massive data, is computationally efficient and applicable to complex real scenarios.

[0006] An intelligent transportation task execution method based on image dehazing, comprising:

[0007] Collecting a road image and calculating the atmospheric light of the road image; wherein, the road image contains haze;

[0008] Screening out the minimum values of multiple color channels of each pixel point in the road image to obtain the minimum color channel;

[0009] Calculating the haze concentration of the road image according to the atmospheric light and the minimum color channel;

[0010] Calculating the transmittance according to the haze concentration;

[0011] Inputting the transmittance and the atmospheric light into an atmospheric scattering model for dehazing to obtain a dehazed image;

[0012] Inputting the dehazed image into an intelligent transportation system to execute an intelligent transportation task.

[0013] In a preferred technical solution of the present invention, the calculating the atmospheric light of the road image includes:

[0014] Performing minimum value filtering on the three color channels of the road image respectively to obtain a minimum value filtered image;

[0015] Screening out the maximum value from each color channel of the minimum value filtered image to obtain the atmospheric light.

[0016] In a preferred technical solution of the present invention, the screening out the minimum values of multiple color channels of each pixel point in the road image to obtain the minimum color channel includes:

[0017] Sequentially taking each pixel point in the road image as the current pixel point;

[0018] Taking the minimum value of the three color channels of the current pixel point as the pixel value of the current pixel point;

[0019] Detecting whether all pixel points in the road image have been traversed, and if so, obtaining the minimum color channel.

[0020] In a preferred technical solution of the present invention, calculating the haze concentration of the road image according to the atmospheric light and the minimum color channel includes:

[0021] Calculating the haze concentration of each pixel point in the road image according to the following formula:

[0022]

[0023] where θ(x, y) is the haze concentration of the pixel point at the x-th row and y-th column in the road image, A is the atmospheric light, and I min (x, y) is the pixel value of the pixel point at the x-th row and y-th column in the minimum color channel.

[0024] In a preferred technical solution of the present invention, calculating the transmittance according to the haze concentration includes:

[0025] Calculating the transmittance according to the following formula:

[0026] t(x, y) = (1 - θ(x, y)) γ(x,y) ;

[0027] where t(x, y) is the transmittance of the pixel point at the x-th row and y-th column in the road image, and γ(x, y) is the correction coefficient of the pixel point at the x-th row and y-th column in the road image.

[0028] In a preferred technical solution of the present invention, inputting the transmittance and the atmospheric light into an atmospheric scattering model for haze removal to obtain a haze-removed image includes:

[0029] Calculating the pixel value of each pixel point of the haze-removed image according to the following formula:

[0030]

[0031] where ρ(x, y) is the pixel value of the pixel point at the x-th row and y-th column of the haze-removed image, A is the atmospheric light, t(x, y) is the transmittance of the pixel point at the x-th row and y-th column of the road image, and I(x, y) is the pixel value of the pixel point at the x-th row and y-th column of the road image.

[0032] In a preferred technical solution of the present invention, inputting the haze-removed image into an intelligent transportation system to perform intelligent transportation tasks includes:

[0033] Extracting the license plate area in the haze-removed image;

[0034] Performing character recognition on the license plate area to obtain the license plate number.

[0035] In a preferred technical solution of the present invention, inputting the haze-removed image into an intelligent transportation system to perform intelligent transportation tasks includes:

[0036] Extract the target in the defogged image;

[0037] Track the target within a preset time period to obtain a target trajectory; wherein, the target trajectory is a vehicle trajectory or a pedestrian trajectory.

[0038] In a preferred technical solution of the present invention, after calculating the transmittance according to the haze concentration, the following steps are further included:

[0039] Select the maximum value between the initial transmittance and the lower limit of the adaptive transmittance to obtain the quantity to be optimized;

[0040] Optimize the quantity to be optimized by using curvature filtering to obtain the optimized transmittance;

[0041] Perform defogging on the road image based on the optimized transmittance to obtain an optimized defogged image.

[0042] The present invention also provides an intelligent transportation task execution device based on image defogging, including:

[0043] An image acquisition module, configured to acquire a road image and calculate the atmospheric light of the road image;

[0044] A minimum color channel screening module, configured to screen out the minimum value of multiple color channels of each pixel point in the road image to obtain a minimum color channel;

[0045] A haze concentration calculation module, configured to calculate the haze concentration of the road image according to the atmospheric light and the minimum color channel;

[0046] A transmittance calculation module, configured to calculate the transmittance according to the haze concentration;

[0047] A defogging module, configured to input the transmittance and the atmospheric light into an atmospheric scattering model for defogging to obtain a defogged image;

[0048] An intelligent transportation task execution module, configured to input the defogged image into an intelligent transportation system to execute an intelligent transportation task.

[0049] The beneficial effects of the present invention are:

[0050] The intelligent transportation task execution method based on image defogging provided by the present invention includes collecting road images and calculating the atmospheric light of the road images, where the road images contain haze. The estimation of atmospheric light is one of the key steps in image defogging. Since atmospheric light is the main reason for the low radiation value of ground objects in remote sensing images, it is necessary to accurately calculate the atmospheric light. The minimum color value of each pixel point in the road image is selected to obtain the minimum color channel, and the minimum color channel can reflect the haze concentration to a certain extent. The haze concentration of the road image is calculated based on the atmospheric light and the minimum color channel. When the minimum color channel approaches 0, the haze concentration approaches 0. When the minimum color channel approaches the atmospheric light, the haze concentration approaches 1. The transmittance is calculated based on the haze concentration. A correction coefficient is introduced to construct a non-linear correction function, and the non-linear correction function is used to map the haze concentration to obtain the transmittance. The transmittance and the atmospheric light are input into the atmospheric scattering model for inversion to remove the haze in the road image and obtain a defogged image. The defogged image is input into the intelligent transportation system to execute intelligent transportation tasks. The intelligent transportation tasks include license plate number recognition and target trajectory tracking, and the target trajectory includes the trajectories of vehicles and pedestrians. The defogging algorithm adopted by the present invention does not rely on massive data, is computationally efficient and applicable to complex real scenes. The intelligent transportation task execution method based on the defogging algorithm can accurately parse license plate characters, realize intelligent monitoring, and accurately track the target trajectory. Description of the Drawings

[0051] Figure 1 is a flowchart of the intelligent transportation task execution method based on image defogging of the present invention;

[0052] Figure 2 is a flowchart of screening the minimum color channel of the present invention;

[0053] Figure 3 is a flowchart of defogging the road image based on the optimized transmittance of the present invention;

[0054] Figure 4 is a structural schematic block diagram of the intelligent transportation task execution device based on image defogging of the present invention.

[0055] Reference Numerals: 10, image acquisition module; 20, minimum color channel screening module; 30, haze concentration calculation module; 40, transmittance calculation module; 50, defogging module; 60, intelligent transportation task execution module. Detailed Embodiments

[0056] The preferred embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the preferred embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0057] Example 1

[0058] As Figure 1 shown, this embodiment provides an intelligent transportation task execution method based on image defogging, including:

[0059] S1: Collect road images and calculate the atmospheric light of the road images; wherein, the road images contain haze.

[0060] S2: Screen out the minimum values of multiple color channels of each pixel point in the road images to obtain the minimum color channel.

[0061] S3: Calculate the haze concentration of the road images according to the atmospheric light and the minimum color channel.

[0062] S4: Calculate the transmittance according to the haze concentration.

[0063] S5: Input the transmittance and the atmospheric light into an atmospheric scattering model for defogging to obtain a defogged image.

[0064] S6: Input the defogged image into an intelligent transportation system to execute an intelligent transportation task.

[0065] Use ITS (Intelligent Transportation System) front-end devices such as surveillance cameras, drones, and / or in-vehicle cameras to collect images and videos of the road environment in real time.

[0066] Evaluate the current haze concentration of the collected road images and decide whether to perform defogging processing. The calculation of the haze concentration can be performed using an inverse transformation method for haze concentration based on prior knowledge. When the calculated haze concentration is low, it is directly output, and when the calculated haze concentration exceeds the haze concentration threshold, defogging processing is performed.

[0067] The calculation of the atmospheric light of the road images includes:

[0068] S12: Perform minimum value filtering on the three color channels of the road images respectively to obtain a minimum value filtered image.

[0069] S13: Screen out the maximum values from each color channel of the minimum value filtered image to obtain the atmospheric light.

[0070] Before step S12, there is also step S11: collecting road images. The road images are RGB images, which include an R channel, a G channel, and a B channel. First, minimum filtering is performed on each color channel of the road images to obtain minimum-filtered images, so as to avoid the interference of image noise. For the minimum-filtered images, the maximum value of each color channel is selected as the atmospheric light. The calculation of the atmospheric light is performed separately for each color channel, and finally a fixed value is obtained for each color channel, that is, the atmospheric light is a fixed value for each color channel.

[0071] Atmospheric light estimation is one of the key steps in image dehazing. Since atmospheric light is the main reason for the low radiation value of ground objects in remote sensing images, it is necessary to accurately calculate the atmospheric light.

[0072] As Figure 2 shown, the obtaining of the minimum color channel by screening out the minimum values of multiple color channels of each pixel point in the road image includes:

[0073] S21: Sequentially take each pixel point in the road image as the current pixel point.

[0074] S22: Take the minimum value among the three color channels of the current pixel point as the pixel value of the current pixel point.

[0075] S23: Detect whether all pixel points in the road image have been traversed. If so, the minimum color channel is obtained.

[0076] The minimum color channel can reflect the haze concentration to a certain extent. The haze concentration of the road image is calculated based on the atmospheric light and the minimum color channel. When the minimum color channel approaches 0, the haze concentration approaches 0. When the minimum color channel approaches the atmospheric light, the haze concentration approaches 1.

[0077] Traverse each pixel point in the road image. Each pixel point has three color channels, namely the R channel, the G channel, and the B channel. For the current pixel point, the minimum value among the three color channels is taken as the minimum color value of the current pixel point.

[0078] The minimum color channel is a grayscale image. After traversing all pixel points in the road image, the minimum color channel is obtained. The size of the minimum color channel is the same as that of the road image, that is, the minimum color values of each pixel point form the minimum color channel, and the pixel value of each pixel point in the minimum color channel is the minimum value among the R channel, the G channel, and the B channel.

[0079] The dehazing algorithm of the present invention can effectively remove haze in complex real - road images, obtain good details, and has a low complexity and high execution efficiency, which can meet the requirements of target recognition and trajectory tracking.

[0080] The intelligent transportation task execution method based on image dehazing provided in this embodiment includes collecting road images and calculating the atmospheric light of the road images, where the road images contain haze. The estimation of atmospheric light is one of the key steps in image dehazing. Since atmospheric light is the main reason for the low radiation value of ground objects in remote sensing images, it is necessary to accurately calculate the atmospheric light. The minimum color value of each pixel point in the road image is screened out to obtain the minimum color channel, and the minimum color channel can reflect the haze concentration to a certain extent. The haze concentration of the road image is calculated based on the atmospheric light and the minimum color channel. When the minimum color channel approaches 0, the haze concentration approaches 0. When the minimum color channel approaches the atmospheric light, the haze concentration approaches 1. The transmittance is calculated according to the haze concentration. A correction coefficient is introduced to construct a non - linear correction function, and the non - linear correction function is used to map the haze concentration to obtain the transmittance. The transmittance and the atmospheric light are input into the atmospheric scattering model for inversion to remove the haze in the road image and obtain a dehazed image. The dehazed image is input into the intelligent transportation system to execute intelligent transportation tasks. The intelligent transportation tasks include license plate number recognition and target trajectory tracking, and the target trajectories include the trajectories of vehicles and pedestrians. The dehazing algorithm adopted by the present invention does not rely on massive data, is computationally efficient and applicable to complex real - world scenarios. The intelligent transportation task execution method based on this dehazing algorithm can accurately parse license plate characters, achieve intelligent monitoring, and accurately track target trajectories.

[0081] Embodiment 2

[0082] The present embodiment provides an intelligent transportation task execution method based on image dehazing. This embodiment only describes the differences from Embodiment 1. The calculating the haze concentration of the road image according to the atmospheric light and the minimum color channel includes:

[0083] Calculating the haze concentration of each pixel point in the road image according to the following formula:

[0084]

[0085] where θ(x,y) is the haze concentration of the pixel point at the x - th row and y - th column in the road image, A is the atmospheric light, and I min (x,y) is the pixel value of the pixel point at the x - th row and y - th column in the minimum color channel.

[0086] The minimum color channel can reflect the haze concentration to a certain extent. The haze concentration is inversely proportional to the atmospheric light, and the haze concentration is directly proportional to the minimum color channel. That is, the greater the value of the atmospheric light, the smaller the haze concentration, and the greater the minimum color channel, the greater the haze concentration.

[0087] Calculating the transmittance according to the haze concentration includes:

[0088] Calculating the transmittance according to the following formula:

[0089] t(x,y) = (1 - θ(x,y)) γ(x,y) ; (2)

[0090] where t(x,y) is the transmittance of the pixel at the x-th row and y-th column in the road image, and γ(x,y) is the correction coefficient of the pixel at the x-th row and y-th column in the road image.

[0091] When the haze concentration θ(x,y) of the pixel at the x-th row and y-th column in the road image increases, the transmittance of the pixel at the x-th row and y-th column in the road image decreases; when the haze concentration θ(x,y) of the pixel at the x-th row and y-th column in the road image decreases, the transmittance of the pixel at the x-th row and y-th column in the road image increases. Through statistical observation, it is determined that the distribution of the transmittance and the inverted haze concentration is similar. Therefore, a correction coefficient is introduced, and a non-linear correction function is used to model the transmittance.

[0092] When the haze concentration is calculated, the correction coefficient needs to be solved. According to the haze concentration and the correction coefficient, the transmittance can be calculated. The present invention uses the following formula to describe the atmospheric scattering model:

[0093] I(x,y) = A.ρ(x,y).t(x,y) + A.(1 - t(x,y)); (3)

[0094] where I(x,y) is the pixel value of the pixel at the x-th row and y-th column of the road image, ρ(x,y) represents the pixel value of the pixel at the x-th row and y-th column of the dehazed image, A is the global atmospheric light, t(x,y) is the transmittance corresponding to the pixel at the x-th row and y-th column of the road image, and. represents the multiplication operation.

[0095] To quickly solve the transmittance, first apply the color averaging operation, that is, calculate the average value of the three channels of the color image, to both sides of formula (3), and transform formula (3) into:

[0096]

[0097] where, represents the average pixel value of the pixel at the x-th row and y-th column of the road image, that is, the average value of the pixel at the x-th row and y-th column of the road image in the three color channels, represents the average pixel value of the pixel at the x-th row and y-th column of the defogged image, that is, the average value of the pixel at the x-th row and y-th column of the defogged image in three color channels. represents the average value of the atmospheric light, that is, the average value of the atmospheric light in three color channels, and '.' represents the multiplication operation.

[0098] Calculate according to the following formula and

[0099]

[0100] where c represents the color channel, r represents the R channel, i.e., the red channel, g represents the G channel, i.e., the green channel, b represents the B channel, i.e., the blue channel, and I c (x, y) represents the pixel value of the pixel at the x-th row and y-th column of the c channel of the road image, and ρ c (x, y) represents the pixel value of the pixel at the x-th row and y-th column of the c channel of the defogged image, and A c (x, y) represents the value of the pixel at the x-th row and y-th column of the c channel of the atmospheric light, represents the average value of the pixel of the atmospheric light at the x-th row and y-th column.

[0101] As an example, if the estimated atmospheric light is expressed as A = {220, 200, 180}, that is, the value of the atmospheric light in the R channel is 220, the value of the atmospheric light in the G channel is 200, and the value of the atmospheric light in the B channel is 180, then the average value of the atmospheric light is 200.

[0102] Substitute formula (2) into formula (4) and rewrite formula (4) into the following form:

[0103]

[0104] Formula (8) is a transcendental equation. Since there is an exponential term in formula (8), it is very difficult to directly solve this equation. To solve this problem, the present invention applies Taylor expansion to solve formula (8) and rewrite formula (8) as:

[0105]

[0106] Formula (9) is a quadratic equation about γ(x, y). Introduce the first intermediate quantity, the second intermediate quantity and the third intermediate quantity, and rewrite formula (9) into the following form:

[0107]

[0108] where is the first intermediate quantity, is the second intermediate quantity, is the third intermediate quantity, and the dot in the present invention represents a multiplication operation.

[0109] Calculate the first intermediate quantity, the second intermediate quantity and the third intermediate quantity according to the following formula:

[0110]

[0111] Through Vieta's theorem, the correction coefficient can be calculated by the following formula:

[0112]

[0113] where γ(x, y) is the correction coefficient of the pixel at the x-th row and y-th column of the road image.

[0114] When the haze concentration of the pixel of the road image, the average pixel value of the pixel of the road image, and the average value of the atmospheric light are known, it is also necessary to calculate the average pixel value of the pixel of the dehazed image.

[0115] Based on formula (4), satisfies the following constraint relationship:

[0116]

[0117] where t(x, y) is greater than or equal to 0 and less than or equal to 1.

[0118] Therefore, in this embodiment, is modeled as:

[0119]

[0120] where τ is the power exponent, and the power exponent τ is a constant greater than 1. In this embodiment, τ is greater than or equal to 1.5 and less than or equal to 2 as an example. represents the average value of the atmospheric light at the pixel of the x-th row and y-th column, that is, the average value of the R channel, G channel, and B channel of the atmospheric light at this point. is the average pixel value of the pixel at the x-th row and y-th column of the road image, that is, the average value of the R channel, G channel, and B channel of the road image at this point.

[0121] Substitute formula (14) into formula (11) to calculate and Substitute and into formula (12) to calculate the correction coefficient γ(x, y), substitute the correction coefficient γ(x, y) into formula (2) to calculate the initial transmittance t initial , calculate the average value of the smallest 5% interval in the initial transmittance to obtain the lower limit of the transmittance t min .

[0122] Lower limit t of transmittance min varies with the road image, that is, different road images correspond to different lower limits of transmittance. The initial transmittance t can be estimated according to the collected road image initial , and then according to the initial transmittance t initial to calculate the lower limit t of transmittance min , t min is dynamically changing. Calculate through formula (14) Then substitute into formula (11) to calculate the third intermediate quantity Substitute the first intermediate quantity The second intermediate quantity and the third intermediate quantity into formula (12) to calculate γ(x, y). Substitute γ(x, y) into formula (2) to calculate the initial transmittance t initial , and calculate the average value within the minimum 5% interval in the initial transmittance t initial to obtain the lower limit t of transmittance min .

[0123] As Figure 3 shown, after calculating the transmittance according to the haze concentration, it further includes:

[0124] S51’: Select the maximum value between the initial transmittance and the adaptive lower limit of transmittance to obtain the quantity to be optimized.

[0125] S52’: Optimize the quantity to be optimized by using curvature filtering to obtain the optimized transmittance.

[0126] S53’: Dehaze the road image based on the optimized transmittance to obtain the optimized dehazed image.

[0127] Since the pixel-based solution method may be affected by image noise, the present invention uses curvature filtering and the lower limit of transmittance to constrain the transmittance, and represents the transmittance as:

[0128] t optimize (x, y) = CF(max(t initial , t min )); (15)

[0129] Wherein, t initial is the initial transmittance, t min is the lower limit of transmittance, CF represents curvature filtering, max represents the maximum value operation, and t optimize (x, y) is the optimized transmittance of the x-th row and y-th column of the road image.

[0130] Each pixel of the road image has a transmittance. If the size of the road image is weight * height, where weight is the width of the road image and height is the height of the road image, then the maximum value operation is performed on the initial transmittance and the transmittance lower limit of each pixel in the road image. After performing this operation on all pixels in the road image, a transmittance matrix to be optimized is obtained. Curvature filtering is performed on the transmittance matrix to be optimized, and the optimized transmittance corresponding to each matrix element is obtained, a total of weight * height optimized transmittances. The weight * height optimized transmittances form an optimized transmittance matrix. Substitute the optimized transmittance, atmospheric light, and road image into the atmospheric scattering model for inversion to obtain a defogged image.

[0131] Curvature filtering filters pixels based on curvature to reduce image noise and enhance image edges. Curvature is a physical quantity that describes the degree of curve bending. In image processing, the curvature is estimated by calculating the degree of bending of the neighborhood around the pixel. Curvature filtering classifies the pixels of the road image according to the size of the curvature, and then takes corresponding filtering operations according to the type of the pixel. The calculation process of curvature filtering is as follows:

[0132] (1) Select a window and slide the window on the image.

[0133] (2) For each pixel in the window, calculate the curvature of its surrounding neighborhood.

[0134] (3) According to the size of the curvature, the pixels are divided into two categories: edge pixels and non-edge pixels.

[0135] (4) For edge pixels, selectively perform enhancement operations to improve the clarity and contrast of the image edges.

[0136] (5) For non-edge pixels, perform smoothing operations to reduce image noise and protect texture details.

[0137] The road image contains haze and image noise, with low clarity and contrast, and is non-differentiable at some points or regions, that is, haze and image noise will affect the smoothness of the image. Curvature filtering is based on a minimum energy optimization model driven by geometric curvature to ensure the geometric continuity of image edges. In addition to haze, the road image generally also includes moving objects such as vehicles and pedestrians. The vehicle body has a streamlined shape, and the contour of the pedestrian consists of multiple curves. Haze is composed of gas molecules and solid molecules and does not have a fixed shape. Therefore, curvature can be used to distinguish dynamic objects and haze in the road image. The initial transmittance t initialIt is calculated based on the road image, and the lower limit of the transmittance is calculated from the initial transmittance. Therefore, curvature filtering is used to optimize the transmittance so that the transmittance can better reflect the curvature changes of dynamic objects in the road image, distinguish dynamic objects and haze in the road image, and use the optimized transmittance for defogging. The optimized defogged image obtained has higher clarity and contrast to accurately remove the haze in the road image, thereby improving the accuracy of license plate recognition and trajectory tracking.

[0138] After calculating the transmittance and the atmospheric light, the defogged image is calculated by inverting the atmospheric scattering model. The calculation formula of the defogged image is as follows:

[0139]

[0140] Among them, ρ(x, y) is the pixel value of the pixel at the x-th row and y-th column of the defogged image, I(x, y) is the pixel value of the pixel at the x-th row and y-th column of the road image, A is the atmospheric light, and t(x, y) is the transmittance corresponding to the pixel at the x-th row and y-th column of the road image.

[0141] In this embodiment, when the haze concentration θ(x, y) of the pixel at the x-th row and y-th column in the road image increases, the transmittance of the pixel at the x-th row and y-th column in the road image decreases; when the haze concentration θ(x, y) of the pixel at the x-th row and y-th column in the road image decreases, the transmittance of the pixel at the x-th row and y-th column in the road image increases. Through statistical observation, it is determined that the distribution of the transmittance and the inverted haze concentration is similar. Therefore, a correction coefficient is introduced, and a non-linear correction function is used to model the transmittance. The defogged image removes the haze in the road image, so the defogged image has higher clarity and contrast. Inputting the defogged image into the intelligent transportation system to execute intelligent transportation tasks can improve the accuracy and efficiency of intelligent transportation tasks.

[0142] Embodiment 3

[0143] This embodiment provides an intelligent transportation task execution method based on image defogging. This embodiment only describes the differences from Embodiment 1. The inputting the defogged image into the intelligent transportation system to execute intelligent transportation tasks includes:

[0144] S61: Extract the license plate area in the defogged image.

[0145] S62: Perform character recognition on the license plate area to obtain the license plate number.

[0146] The dehazed image removes the haze in the road image, enhancing the imaging clarity and contrast. Image segmentation is performed on the dehazed image to obtain multiple image regions. The license plate region is screened out from the multiple image regions by using gradient information projection statistics, wavelet transform or license plate region scanning connection algorithm, and character segmentation is performed on the license plate region to obtain image blocks. The character segmentation method uses a connected component analysis algorithm, a projection method or an edge-based segmentation algorithm. Template matching, statistical feature analysis or a neural network-based character recognition algorithm is used to perform character recognition on the image blocks to obtain the license plate number.

[0147] Inputting the dehazed image into the intelligent transportation system to perform intelligent transportation tasks includes:

[0148] S63: Extract the target in the dehazed image.

[0149] S64: Track the target within a preset time duration to obtain the target trajectory; wherein, the target trajectory is a vehicle trajectory or a pedestrian trajectory.

[0150] Use a Fast-RCNN or YOLOv5 detection model to identify the target in the dehazed image, and then combine multiple frames of dehazed images to track the target to obtain the target trajectory. The target can be dynamic, such as a vehicle or a pedestrian, and the target can also be static, such as a traffic light or a zebra crossing.

[0151] In this embodiment, inputting the dehazed image into the intelligent transportation system to perform intelligent transportation tasks, the intelligent transportation tasks include character recognition and trajectory tracking. Character recognition includes extracting the license plate region in the dehazed image, performing character recognition on the license plate region to obtain the license plate number. Trajectory tracking includes extracting the target in the dehazed image, tracking the target within a preset time duration to obtain the target trajectory, wherein the target trajectory is a vehicle trajectory or a pedestrian trajectory. Use the image dehazing algorithm in Embodiment 1 or Embodiment 2 to dehaze the road image to obtain the dehazed image. The dehazed image has a high resolution and contrast. Based on the dehazed image for character recognition and trajectory tracking can improve the accuracy of character recognition and trajectory tracking.

[0152] Embodiment 4

[0153] As Figure 4 shown, this embodiment provides an intelligent transportation task execution device based on image dehazing, including:

[0154] An image acquisition module 10, configured to acquire a road image and calculate the atmospheric light of the road image;

[0155] A minimum color channel screening module 20, configured to screen out the minimum value of multiple color channels of each pixel point in the road image to obtain the minimum color channel;

[0156] The haze concentration calculation module 30 is configured to calculate the haze concentration of the road image according to the atmospheric light and the minimum color channel;

[0157] The transmittance calculation module 40 is configured to calculate the transmittance according to the haze concentration;

[0158] The defogging module 50 is configured to input the transmittance and the atmospheric light into an atmospheric scattering model for defogging to obtain a defogged image;

[0159] The intelligent transportation task execution module 60 is configured to input the defogged image into an intelligent transportation system to execute an intelligent transportation task.

[0160] The intelligent transportation task execution device based on image defogging in this embodiment is used to execute the intelligent transportation task execution method based on image defogging in any one of Embodiments 1 - 3.

[0161] This embodiment also provides a computer device, which may be a server. Among them, the computer device includes a processor, a memory, a network interface, and a database connected through a system bus. The processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection.

[0162] This embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the intelligent transportation task execution method based on image defogging described in any one of Embodiments 1 - 3. It can be understood that the computer-readable storage medium in this embodiment may be a volatile readable storage medium or a non-volatile readable storage medium.

[0163] It should be noted that in this article, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, apparatus, article or method including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, apparatus, article or method. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, apparatus, article or method including that element.

[0164] The above are only the preferred embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present application.

Claims

1. An intelligent transportation task execution method based on image dehazing, characterized in that, Including: Collecting a road image and calculating the atmospheric light of the road image; wherein, the road image contains haze. Selecting the minimum value of multiple color channels of each pixel point in the road image to obtain a minimum color channel. Calculating the haze concentration of the road image according to the atmospheric light and the minimum color channel. Calculating the transmittance according to the haze concentration. Inputting the transmittance and the atmospheric light into an atmospheric scattering model for haze removal to obtain a haze-removed image. Inputting the haze-removed image into an intelligent transportation system to perform intelligent transportation tasks.

2. The intelligent transportation task execution method based on image defogging according to claim 1, wherein, The calculating of the atmospheric light of the road image includes: Performing minimum value filtering on three color channels of the road image respectively to obtain a minimum value filtered image. Selecting the maximum value from each color channel of the minimum value filtered image to obtain the atmospheric light.

3. The intelligent transportation task execution method based on image dehazing according to claim 1, wherein The selecting of the minimum value of multiple color channels of each pixel point in the road image to obtain a minimum color channel includes: Sequentially taking each pixel point in the road image as the current pixel point. Taking the minimum value of the three color channels of the current pixel point as the pixel value of the current pixel point. Detecting whether all pixel points in the road image have been traversed. If so, the minimum color channel is obtained.

4. The intelligent transportation task execution method based on image dehazing according to claim 1, characterized in that The calculating of the haze concentration of the road image according to the atmospheric light and the minimum color channel includes: Calculating the haze concentration of each pixel point in the road image according to the following formula: Among them, θ(x, y) is the haze concentration of the pixel at the x-th row and y-th column in the road image, A is the atmospheric light, and I min (x, y) is the pixel value of the pixel at the x-th row and y-th column in the minimum color channel.

5. The intelligent transportation task execution method based on image dehazing according to claim 4, wherein, The calculating of the transmittance according to the haze concentration includes: Calculating the transmittance according to the following formula: t(x,y) = (1 - θ(x,y)) γ(x,y) ; Wherein, t(x, y) is the transmittance of the pixel point at the x-th row and y-th column in the road image, and γ(x, y) is the correction coefficient of the pixel point at the x-th row and y-th column in the road image.

6. The intelligent transportation task execution method based on image dehazing according to claim 1, wherein The inputting of the transmittance and the atmospheric light into an atmospheric scattering model for haze removal to obtain a haze-removed image includes: Calculating the pixel value of each pixel point of the haze-removed image according to the following formula: Wherein, ρ(x, y) is the pixel value of the pixel point at the x-th row and y-th column of the haze-removed image, A is the atmospheric light, t(x, y) is the transmittance of the pixel point at the x-th row and y-th column of the road image, and I(x, y) is the pixel value of the pixel point at the x-th row and y-th column of the road image.

7. The method for performing an intelligent transportation task based on image defogging according to claim 1, wherein The inputting of the haze-removed image into an intelligent transportation system to perform intelligent transportation tasks includes: Extracting the license plate area in the haze-removed image. Performing character recognition on the license plate area to obtain the license plate number.

8. The intelligent transportation task execution method based on image dehazing according to claim 1, characterized in that The inputting of the haze-removed image into an intelligent transportation system to perform intelligent transportation tasks includes: Extracting the target in the haze-removed image. Tracking the target within a preset time period to obtain a target trajectory; wherein, the target trajectory is a vehicle trajectory or a pedestrian trajectory.

9. The intelligent transportation task execution method based on image dehazing according to claim 1, characterized in that, After the calculating of the transmittance according to the haze concentration, it further includes: Selecting the maximum value between the initial transmittance and the adaptive transmittance lower limit to obtain a quantity to be optimized. Optimizing the quantity to be optimized by using curvature filtering to obtain an optimized transmittance. Performing haze removal on the road image based on the optimized transmittance to obtain an optimized haze-removed image.

10. An intelligent transportation task execution device based on image dehazing, characterized in that, Including: An image acquisition module for collecting a road image and calculating the atmospheric light of the road image. The minimum color channel screening module is used to screen out the minimum value of multiple color channels of each pixel point in the road image to obtain the minimum color channel; The haze concentration calculation module is used to calculate the haze concentration of the road image according to the atmospheric light and the minimum color channel; The transmittance calculation module is used to calculate the transmittance according to the haze concentration; The defogging module is used to input the transmittance and the atmospheric light into the atmospheric scattering model for defogging to obtain a defogged image; The intelligent transportation task execution module is used to input the defogged image into the intelligent transportation system to execute intelligent transportation tasks.

Images