Electronic rearview mirror image defogging method based on image information integration

Abstract

The invention discloses an electronic rearview mirror image defogging method based on image information integration, and the method comprises the steps: S101, obtaining real-time ambient light data, and dynamically adjusting the contrast and brightness of an image according to the data; s102, generating an optimized basic image based on the adjusted image parameters; s103, adjusting parameters of an image edge enhancement algorithm according to real-time meteorological conditions; and S104, applying the adjusted parameters to the basic image to output a clear defogged image, and dynamically adjusting the contrast and brightness of the image according to the change of ambient light to solve the problem of imaging blurring. And adjusting parameters of an image edge enhancement algorithm according to real-time meteorological conditions so as to solve the problem that the image boundary is not clear in foggy days.

Description

Technical Field

[0001] The present application relates to the technical field of electronic rearview mirror image processing, and in particular to an electronic rearview mirror image defogging method based on image information integration. Background Art

[0002] The electronic rearview mirror image defogging method with image information integration is a technology designed to improve the image quality of the electronic rearview mirror under adverse weather conditions, especially in foggy days. It integrates a variety of advanced image processing technologies to effectively reduce or remove the negative impact of fog on the field of vision on the basis of capturing the external environment image, so that the image can be restored to a clearer and more realistic state.

[0003] There are two key technical issues it faces: the first is how to dynamically adjust the contrast and brightness of the image according to the changes in ambient light to solve the problem of blurred imaging. This is because the different intensity of external light will greatly affect the color saturation and brightness of the image captured by the camera. If the image is overexposed or underexposed under different lighting conditions, the visual information will be damaged; The second is to adjust the parameters of the image edge enhancement algorithm according to the real-time meteorological conditions to solve the problem of unclear image boundaries in low visibility conditions such as foggy days. Because when the fog is thick, the scattering effect of particulate matter in the air on light increases significantly, which makes the outline of objects blurred and difficult to recognize. Therefore, it is necessary to adaptively correct the corresponding settings in the image processing model according to the actual climate conditions to ensure that the output results have higher resolution and sharpness. Summary of the invention

[0004] In view of this, an embodiment of the present disclosure provides an electronic rearview mirror image defogging method based on image information integration, which at least partially solves the problems existing in the above-mentioned prior art.

[0005] An electronic rearview mirror image defogging method based on image information integration, comprising: S101: Acquire real-time ambient light data, and dynamically adjust the contrast and brightness of the image according to the data; S102: generating an optimized basic image based on the adjusted image parameters; S103: adjusting parameters of the image edge enhancement algorithm according to real-time meteorological conditions; S104: Applying the adjusted parameters to the base image to output a clear defogging image.

[0006] In a specific embodiment, it further includes: determining an initial brightness enhancement coefficient A0 based on the vehicle speed and the camera viewing angle range, wherein the faster the vehicle speed is and the larger the camera viewing angle range is, the greater the value of A0 is; According to the real-time ambient light intensity E, the brightness adjustment formula is adopted: A = A0 * (1 + tanh(E / (Es * Ec))); where Es is the ambient light intensity threshold, Ec is the ambient light intensity center offset, and tanh is the hyperbolic tangent function, which is used to map the input value to the interval (-1, 1); Dynamically update the photosensitivity G of the camera sensor: G = Gbase + A * ΔG, where Gbase is the base sensitivity and ΔG is the photosensitivity gain change; If the current ambient light intensity E is greater than the preset maximum light intensity Em, then set A = 1 and keep G unchanged to prevent overexposure.

[0007] In a specific embodiment, it further includes: initially selecting a suitable fog segmentation strategy S0 based on the precipitation R in the weather information; After grayscale processing the image, use the Sobel operator to calculate the gradient. If the average gradient M < Th, use high-pass filtering to enhance the image; Calculate the local contrast enhancement factor Cy: Cy = exp(α * M^2 / σs^2); where α and σs are the contrast influence coefficient and standard deviation respectively, and exp is the exponential function; Fuse the original image and the filtered image by weighted average of pixels: W = (Img_original + C * Img_filter) / 2.

[0008] In a specific embodiment, it further includes: adjusting the sharpening intensity K0 = V / Vmax according to the vehicle speed V according to a preset rule. The faster the vehicle speed, the greater the sharpening intensity; Pre-adjust the contrast parameter of the image for the predicted humidity H in the next few frames, and correct it using the following formula: B = β(H50), where β is the humidity change response factor; Real-time detect and feedback the current visual visibility VD, and set the visibility compensation coefficient γ(VD) = 7 + 1 * sqrt(abs(log(VD / 10))). When VD reaches the set threshold Tvd, reduce the K weight to protect the user's eye health; Loop the above operations at different time intervals to ensure the stability and real-time performance of the video stream.

[0009] In a specific embodiment, it further includes: obtaining the average light intensity Ep_avg and standard deviation σEp within the previous second; When it is detected that the ambient light fluctuates rapidly, that is, abs(Ep_current - Ep_previous) > τ * σEp, where τ is the sensitivity threshold, and perform transient adaptive adjustment; Perform color balance optimization ColorAdj, and use the Lab color space to perform non-linear mapping relationships for separation and conversion of L, a*, b*; Use the formula F(C)=θ*C²+(1-θ)*C to adjust the interaction between channels in the color space, ensuring the best final image quality, where C represents the intensity of any single color channel and θ is the balance adjustment parameter.

[0010] In a specific embodiment, it further includes: precisely selecting the degree e0 of image edge enhancement according to the current atmospheric humidity HR; When HR < hr_limit, execute the low humidity mode: refine the target edge through a bilateral filter; If high humidity triggers the high humidity protection mechanism: F_HighHR = w_α*max(exp(μt*t), w_ω*cos(π*f*d)), t∈[t1, t2] Apply an improved bilateral filtering function to process the image I = F(I, f, d), making the vehicle edges in the fog environment more clearly visible. f is the spatial filtering radius, d is the spatial distance, μt, wt control the smoothing effect in the time dimension, w_ω controls the frequency weight, and hr_limit represents the critical humidity limit.

[0011] In a specific embodiment, it further includes: introducing an intelligent environment assessment model to predict possible future environmental changes and making adaptive configuration changes in advance; Define a comprehensive evaluation index Zeta to judge the effectiveness of image processing: Zeta(i)=Σ[Wi*log(δi*(Cy - Di))^(η)], where Cy is the statistical attribute value of the processed image such as mean square error, etc., and Di is the corresponding ideal value; Use the maximum Zeta value obtained under specific conditions as the guiding principle to determine the direction of the image defogging strategy in the next cycle; If the calculation results show a significant decrease for two or more consecutive times, start the reset learning stage to re-learn the optimal parameter sets Wi, η, etc.

[0012] In a specific embodiment, it further includes: comprehensively considering the changes in the illumination angle and direction, automatically correcting the influence degree factor_dirction of the direct light source. This can be achieved by establishing an illumination model, calculating the influence of the direct light source on the image according to the illumination angle and direction, and making corresponding corrections; Divide the collected data into two parts, one part of the training samples is used to construct a short-term trend prediction engine model_shortTermPred, and the other part is the verification data testdata_valiation; Use cross-validation to determine the best hyperparameters of the model; sets_optimParam, such as learningRate, nHiddenLayers, etc.; Enable special mode when encountering extreme weather phenomena, that is, if WeatherExt == True: useExtremeProcessingAlgo().

[0013] In a specific implementation, it further includes: establishing a personalized memory bank Memory_bank in combination with historical driving scenarios, so that users can retrieve previous successful cases for reference and comparison every time they encounter similar situations; Set the corresponding priority prio according to the terrain class terrainClass where the vehicle is located to arrange the subsequent steps; Calculate the number of effective elements N_effective in each frame image, and perform additional amplification or enhancement for those less than threshold_num to avoid missing important details; Context-related information contextInfo is introduced to assist in identifying the background noise interference source ContextFilter(contextInfo)=contextImportanceWeightedMean(contextStrength).

[0014] In a specific implementation, it further includes: using the deep learning resources provided by the cloud data center to continuously improve the denoising performance of the local port NoiseRedutionPerformance_improvememt (CloudResrouce), monitoring the network connection quality networkStableLevel to ensure stable transmission efficiency, and when there is packet loss or unsatisfactory delay NetworkQuality<=low_threshold, starting the emergency offline operation plan ContingencyPlanofflineMode(). The denoising performance improvement here can denoise the local image through the deep learning model in the cloud, and then feed back the processing results to the local port; Check whether the latest version of the software has any unfixed vulnerabilities BugReport_latest, and push the patch package PatchDeployment() in time; In a specific implementation, it further includes: analyzing the driver's habitual actions and habitual preferences habit_patterns to select an adaptive image presentation mode Presentation_mode (habit_pattern).

[0015] Regularly maintain an online knowledge graph KnowledgeGraph_online to collect new information about road conditions NewRoadInfo and update it in a timely manner; Determine whether to activate certain features feature_activation_status according to the latest traffic regulations TrafficRegulations and real-time weather conditions RealTimeWeaherSituation; In response to emergency braking or other emergencies, EmergencyResponse() quickly captures and saves the accident scene for later review InCydentReviewProcess().

[0016] The disclosed embodiment provides an electronic rearview mirror image defogging method based on image information integration, which obtains the original image collected by an external camera and then enters the preprocessing stage. In this process, the light sensor and the like are used to sense the ambient light intensity. When the light conditions are good, the system will use a relatively mild filtering algorithm to reduce image noise and protect the original details of the image; and when it is detected that the environment is under weak light or variable light conditions, it can intelligently compensate the image and dynamically improve the contrast and brightness, thereby solving the problem of image blur by dynamically adjusting the contrast and brightness of the image according to the change of ambient light; By completing the image post-processing process: such as removing moiré or correcting geometric distortion, all the above steps are automatically controlled under the premise of ensuring safety and reliability, and fully consider the high-speed operation characteristics of the vehicle, which has extremely high delay requirements. Therefore, the entire algorithm runs very efficiently and quickly with reasonable resource consumption, which is suitable for the needs of modern automotive applications. This not only improves the visual effect in rainy, snowy and foggy conditions, but also improves the driving safety factor. It can adjust the parameters of the image edge enhancement algorithm according to real-time meteorological conditions to solve the problem of unclear image boundaries in foggy days. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the exemplary implementation methods of the embodiments of the present disclosure, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the embodiments of the present disclosure and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 The present invention is a flow chart of an electronic rearview mirror image defogging method based on image information integration. DETAILED DESCRIPTION

[0019] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0020] Next, the specific implementation steps of the electronic rearview mirror image defogging method based on image information integration of the present invention are described with reference to the accompanying drawings, and the process of how to solve the image contrast and brightness blurring caused by ambient light changes and the unclear edge of foggy images. The method includes multiple core links, aiming to improve the imaging quality of the electronic rearview mirror under different weather conditions through advanced image processing algorithms, thereby ensuring driving safety.

[0021] First, collect images of actual road scenes with atmospheric attenuation factors. In actual operation, a high-resolution camera is installed on the rear side of the vehicle body to capture the scene behind the vehicle in real time and obtain the initial grayscale image and RGB color image. To ensure the integrity and reliability of the data, these input frames need to undergo a series of pre-filtering processes, such as histogram equalization to eliminate the noise introduced by the inherent defects of optical devices, and then obtain one or more clear, clean images that can truly reflect the characteristics of the surrounding environment for further calculation.

[0022] After that, multi-scale decomposition is performed to separate the atmospheric light components, capture the hidden structural features from the detail level, and estimate the global haze intensity coefficient value in order to compensate for the effective texture information lost due to the scattering effect in the later stage, and initially establish a map of transmittance, which provides a basis for the next step of accurately restoring the polluted target, and on this basis, build a performance model framework with robust performance in complex environments. For example, on a cloudy day, when the sensor detects a low amount of light and determines that the current concentration is in the range of light haze, a more relaxed threshold condition can be adopted to reduce the risk of excessive suppression of high-frequency signals, thereby avoiding excessive sharpening and distortion that affects the overall visual experience.

[0023] Next, in order to solve the problem of dynamically adjusting the contrast and brightness according to the changes in ambient light mentioned above, we adopt a local adaptive dual enhancement strategy. On the one hand, we combine the exposure balance principle to adjust the differences in pixel density distribution between each sub-window; on the other hand, we use the gamma correction function mapping method to effectively extend the grayscale range to expand the visibility in the dark area, narrow the excessive gap between the two, and ensure the output result. It can present bright and saturated colors while having a high enough resolution, so that even in low light, it can still maintain a high resolution and reduce the probability of blurring.

[0024] To solve the problem of blurred imaging, an intelligent prediction module is introduced. By mining and refining the historical statistical data accumulated in the early stage, an empirical law model is summarized to guide the current state migration process to select the most suitable parameter combination form, and automatically match the best effect setting to achieve the purpose of real-time synchronization. In one embodiment, the system will monitor various lighting conditions such as sunny, evening or night outside, and autonomously learn the best response mechanism corresponding to each situation through the embedded learning algorithm. Once triggered, the working mode will be switched immediately to quickly complete the relevant correction actions to ensure that high-quality images are always presented to the driver.

[0025] Finally, a complete set of automated evaluation systems was studied for the problem of parameter regulation of edge enhancement algorithms under real-time meteorological conditions. The main idea is to first define several key quantitative indicators based on physical characteristics, such as the uniformity of the contrast boundary slope, and then use deep convolutional neural network technology to build a simulation environment to simulate all typical conditions that may occur in nature, and finally determine the different disturbance levels caused by various types of cloud, fog, rainfall and other factors. Based on this, a corresponding fine-tuning rule library is established to facilitate the program, directly query and reference the instant online correction weight factor until it fully matches the expected output characteristics, completely improving the problems of stiff edge transitions that are common in the original system, and greatly improving the accuracy of target recognition.

[0026] In general, in order to cope with the ever-changing weather conditions in reality, this new image information integrated interactive processing platform integrates a variety of advanced technical means, comprehensively utilizes statistical prior theoretical knowledge to carefully construct a multi-level progressive architecture, and realizes a functional closed loop that fully covers the entire business logic chain. It can always maintain a stable high-performance level under various extreme working conditions, and meet the higher requirements for the intelligence of vehicle auxiliary equipment in the field of modern smart transportation. Specifically, this solution not only significantly reduces the accident rate of driving line of sight interference caused by adverse weather, but also greatly improves the comfort satisfaction of drivers and passengers and promotes more humanized human-computer interaction.

[0027] Next, the present invention is described to obtain real-time ambient light data and dynamically adjust the contrast and brightness of the image based on the data. The whole process is divided into several steps, each of which has a clear execution function to ensure accurate adjustment, optimize image quality and user viewing experience. First, the system needs to collect light intensity information from external sensors. The external sensor here usually refers to one or more photodetectors placed around the electronic rearview mirror device for measuring ambient light intensity (in Lux, 0 to 100,000 Lux). The goal of this operation is to capture the most immediate changes in brightness levels that match the real world. For example, when the electronic rearview mirror is exposed to strong light that suddenly appears at the exit of a tunnel, the component can quickly sense the information of a sharp increase in ambient light.

[0028] Next, after collecting accurate data, the algorithm enters the next step of preprocessing the original video frame. That is, the obtained illumination value is converted into an input parameter suitable for the contrast enhancement calculation model. In order to complete this step, in one embodiment, the ambient light intensity L is linearly transformed by the formula L=a*L+b to generate a new ambient light intensity L. In the above expression, a is a proportional constant, and its value is between [0.5, 2] and the optimal choice depends on the needs of the actual scene. For example, it is recommended to set it to 1 in the car rearview camera system; b is a bias, which usually takes an integer value in the range of [-50, +50], and setting -30 may be more appropriate for reducing night noise. The purpose of this formula is to scale and calibrate the information captured from the physical world to ensure consistency with the subsequent contrast enhancement logic.

[0029] After the above preprocessing steps, the system is ready to automatically correct the image features based on the converted L. On the one hand, the overall picture brightness B is adjusted. Specifically, if the current ambient illuminance is found to be too bright, the average brightness of the output image is gradually reduced according to the inverse proportional function B=(K / (L+E)) to avoid overexposure. In this relationship, K represents the upper limit of the saturation point (the maximum recommended value is 650), and E represents the minimum normal number (approximately set to 1) that prevents the denominator from approaching zero, thereby avoiding the destruction of display stability due to abnormal increase of calculation results in extreme cases.

[0030] On the other hand, there is the dynamic adjustment of contrast Cy. Similarly, a certain mapping relationship is used to achieve this. Assume that the formula Cy=max(c*(LD), Cy_min), where c represents the gain coefficient, and its floating domain is set to [0.75, 1.33]. It is preferred to adaptively select smaller values ​​such as 0.9 on city streets for different road conditions to maintain a stable visual transition. D is used as a threshold control to offset the influence of atypical weather factors (such as dense fog, heavy rain, etc.). In this context, it can be taken as about 45. C_min ensures that the final contrast is not lower than a certain basic standard (set to 50) to maintain a basically recognizable picture even in severe weather conditions.

[0031] For example, in an actual application scenario, the electronic rearview mirror image defogging method based on image information integration involves the situation of a vehicle driving out of an underground garage during a rainy night. In such a scenario, due to the rapid increase in external light and the presence of a large number of reflective interference sources, the uncorrected camera images often appear gray or too pale. However, after applying the above technical improvements, thanks to the intelligent perception of external light conditions and timely fine-tuning of display parameters, the driver can not only obtain clear and visible feedback on the rear and both sides of the vehicle, but also reduce the possibility of eye fatigue during long-term viewing. Through dynamic optimization post-processing, the main factors affecting visual safety in various adverse optical environments can be effectively overcome.

[0032] Next, the invention is described to generate an optimized basic image based on the adjusted image parameters; in this process, the following steps are included: collecting image data, determining and adjusting image parameters, applying these parameters to the image, and finally evaluating and optimizing the quality of the basic image.

[0033] Specifically: First, in the process of acquiring images, a collection of images of the original electronic rearview mirror scene is captured from multiple sources such as cameras to build a raw image library for processing. In this example, it is assumed that the image is collected from a moving car on a busy road to ensure that the image data can fully represent the real situation that may be encountered.

[0034] After that, the necessary image parameters are selected and adjusted according to specific needs. This includes contrast, brightness, color temperature, and haze removal coefficient. In this link, special attention is paid to how to set image parameter values ​​that are closely related to weather or visibility. For example, for electronic rearview mirror image defogging, an important adjustment is to finely control the variable of transmittance t (range 0 to 1, when t is closer to 1, it means less fog); and atmospheric light value A (dynamically changes according to the actual environment). In order to achieve the ideal effect, the ideal range of transmittance t is set to about 0.5 to 0.8; atmospheric light A needs to obtain the most suitable value range for the current scene based on real-life testing to ensure that the set parameters can maximize the improvement of image visibility and color accuracy after being disturbed by fog.

[0035] In one embodiment, a mathematical formula I=frac{IA}{t}+A is used, where I is the processed image intensity, I is the original image input, A refers to the ambient atmospheric light, and t is the adjusted transmission rate parameter mentioned above. The principle of this formula is to try to restore clear scene information without fog and reduce the problem of visual distortion caused by dense fog.

[0036] Specifically, after completing the above calculations and adjustments, the adjusted parameters are then applied to the initially collected basic image. By executing corresponding algorithms or technical means such as histogram equalization and sharpness enhancement, the optimized basic image is made closer to the real picture quality intuitively felt by the naked eye of humans, while improving the efficiency of identifying objects in the image.

[0037] Then, we enter the quality assessment and optimization phase, using pre-defined criteria such as root mean square error (RMSE) or subjective scoring methods to measure the difference between the image before and after improvement, and make targeted adjustments to continue iterating until the expected effect is achieved. For example, in an actual test case, if the optimized image fails to effectively eliminate the problem of reduced road edge recognition in heavy fog, we may return to fine-tune the A or t value until the most suitable combination is found.

[0038] Next, the method of adjusting the parameters of the image edge enhancement algorithm according to the real-time meteorological conditions of the present invention is described.

[0039] First, the steps include meteorological condition detection, analyzing real-time data to determine the fog or haze density level, adjusting the filter parameters and intensity coefficients based on the analysis results, and finally improving the image processing effect by optimizing the parameters. Meteorological condition detection refers to the use of specific equipment such as meteorological sensors to obtain key data such as temperature, humidity, and atmospheric visibility. This step is crucial to the entire process because an accurate understanding of the external situation lays a solid foundation for subsequent calculations. This information not only affects the final visual quality but also determines the subsequent parameter setting strategy. For example, in high humidity conditions, water vapor condensation is easy to form, resulting in a hazy image. At this time, appropriate adjustments must be made to improve imaging clarity.

[0040] Analyzing real-time data means extracting useful features from the data set obtained in the early stage and calculating the haze value HazeLevel. Specifically, the HazeLevel formula is defined as: HazeLevel=αtimeshumidity+βtimesVisibility^γ, where humidity represents relative humidity, with a value range of [0, 100]; Visibility represents visibility (km), and the value is greater than 0 in normal environment. α and β are proportional coefficients, usually set to constants between 0 and 1, and γ is a power term, which is set to 0.5 by default. This formula is established to weightedly reflect the impact of humidity and visible distance on the image. The greater the humidity and the lower the visibility, the greater the HazeLevel. This indicator can evaluate the severity of the actual situation in order to select the corresponding level of response measures.

[0041] Then, the weight w of the edge detection operator (such as Sobel) and the sigmaColor and sigmaSpace values ​​of the bilateral filter are fine-tuned according to different haze conditions. The weight w belongs to the range [0-1]. When facing thick fog, it will increase to nearly 1 to make the image boundary more eye-catching, which helps the driver to see the details around the vehicle. For σColor and σSpace, the two control the tolerance limits of color and smoothness respectively. In a misty or lightly polluted environment, σColor takes a smaller positive value (such as 2-5), while in heavy weather, σColor is correspondingly expanded to ensure that both the noise of the spots can be reduced and the outline of the object is not lost; similarly, σSpace also has a similar change pattern, which is roughly maintained between 2-8 to maintain the best filtering performance.

[0042] The last step aims to perform corrected rendering on the initial input photo using the adjusted new parameters, and repeatedly compare with the original state to verify the effectiveness of the improvement to ensure the plan is effective, continuously approaching the ideal goal so as to provide a stable and reliable auxiliary tool to serve the actual needs. In one embodiment, assume that an electronic rearview mirror image defogging method based on image information integration faces the test of variable outdoor factors. When the monitoring shows that the current condition is covered by haze, the relevant coefficients are adjusted in a timely manner according to the above logic to make the picture restoration degree higher and thus enhance safety. At the same time, through the dual tests of subjective human eye evaluation and objective numerical comparison, it is proved that the algorithm after dynamic adjustment does improve the visibility and detail presentation ability within the monitoring field of view in complex climates.

[0043] Next, the application of the adjusted parameters to the base image to output a clear defogged image of the present invention will be described. This process is divided into the following steps: obtaining the base image, calculating the atmospheric light value and the transmission map, performing defogging processing according to the optimization formula, and applying the parameter optimization result.

[0044] Specifically, obtaining the base image involves the electronic rearview mirror sensor capturing the initial scene information to form a raw picture. This image may have its visual quality degraded due to factors such as fog. In this example, assume there is a method based on image information integration. First, a road photo with thick fog is collected from the rearview camera equipped on the vehicle as the object to be processed. For example, in one embodiment, this step provides data support for the subsequent algorithm.

[0045] Subsequently, after obtaining the original foggy image, the second step is entered, which is to estimate the atmospheric light A and solve the transmission map t(x). The former is the brightness of the light source component with the most severe scattering in the air, while the latter represents the proportional coefficient of the object passing through a certain thickness of the medium to reach the camera. Generally, 0 < t < 1, t = 0 means invisible, and t = 1 refers to the extreme case where there is no scattering phenomenon; and the atmospheric light intensity value should preferably be selected as the pixel value with the highest gray level in a non-sky area to exclude the misleading influence caused by the sky. For example, in this embodiment, by automatically selecting multiple dark color blocks (excluding the sky area), and finding the maximum brightness for each of the RGB three channels within these blocks to estimate a globally applicable atmospheric light A, and setting the range 0.5 < A < 2.0, and the best value is about 1.3 for generality considerations under different conditions. In addition, deep learning technology or empirical formulas are used to predict the actual transmission ratio factors at each point to form a complete mapping chart, so that even in a large-scale diffuse natural environment, it can ensure that each pixel corresponds to a reasonable transmission attribute, thus ensuring that the accuracy of the whole process is not affected by weather changes. Among them, the dark channel prior method or neural network and other models are used to establish an accurate physical model to estimate the transmission rate function, ensuring good adaptability to various complex road conditions.

[0046] After obtaining the above two key elements, the next step is to remove the haziness in the picture and improve the overall visual clarity according to the pre-built and adjusted algorithm expression. Here, the restoration formula I(j)=((Rc(j)-Ac)) / max(βt(i), τ))+Ac is introduced. In this formula, I(j) indicates the color vector form of any point j in the target image that is finally presented, and Rc(j) indicates the color vector form of the corresponding position received by the mixed observation input port. At the same time, Ac represents the atmospheric light constant mentioned above; max(βt(j), τ) means that in order to prevent the denominator from approaching zero, the minimum guarantee line τ is set and the attenuation coefficient β is multiplied to ensure that there will be no result output that deviates too much from the actual situation. Here, 0.1≤τ≤0.4 and 0<β≤1 are generally set. Ideally, β=0.9 and τ=0.1 can effectively suppress the detail distortion caused by excessive magnification. Specifically, in this implementation case, the already calculated atmospheric light A and transmission matrix are used to complete the entire image quality conversion operation.

[0047] The last step is to feed the adjusted parameters back to the system to check again whether they are correct and then generate the final version. In this process, the difference between the same scene before and after the experiment will be repeatedly compared until it is satisfactory. For example, in this implementation example, the entire process is completed by continuously optimizing and adjusting until the rearview mirror image on the display screen reaches the best state that matches the actual situation well without losing any details.

[0048] The present invention provides an electronic rearview mirror image defogging method based on image information integration, comprising: The method first obtains the original image captured by the external camera. Then, it enters the preprocessing stage, during which the ambient light intensity is sensed using light sensors. When the light conditions are good, the system will use a relatively mild filtering algorithm to reduce image noise and protect the original details of the image; when it detects that the environment is low light or variable light conditions, it can intelligently compensate the image to dynamically improve the contrast and brightness. For example, at night or in a dim tunnel, the algorithm can appropriately increase the dynamic range of the image histogram to highlight important information and make the driving field of view clearer and more clear. Otherwise, the saturation gain is reduced to ensure color accuracy and avoid overexposure.

[0049] Subsequently, after determining the current weather conditions (such as sunny or foggy), this invention makes improvements for the special conditions in the haze environment. Since the image boundary is unclear due to the scattering effect caused by the fog, an adaptive edge enhancement technology is used. That is, relying on real-time meteorological data, such as air quality and climate parameters provided by humidity sensors, PM2.5 detectors, etc., the high-pass factor value in the edge extraction filter is dynamically adjusted to ensure ideal clarity under any meteorological conditions. At the same time, the adjustment intensity will be changed accordingly for different concentrations of haze effects. The soft mode can meet the needs under mild haze pollution; if there is a severe haze invasion, the enhanced processing mechanism is turned on, and through multi-scale convolution kernel iterative operations, the real shape and boundary lines of the target object are gradually restored from coarse granularity to fine granularity, thereby effectively improving the driver's recognition of the scenery outside the car under harsh conditions.

[0050] Finally, the image post-processing process is completed: such as removing moiré or correcting geometric distortion. All of the above steps are automated under the premise of ensuring safety and reliability, and fully consider the high latency requirements of the high-speed operation characteristics of the vehicle. Therefore, the entire algorithm runs very efficiently and quickly with reasonable resource consumption, which is suitable for the needs of modern automotive applications. This not only improves the visual effect in rain, snow and fog, but also improves the driving safety factor. In addition, this solution can also be closely integrated with autonomous driving technology, which helps to play a key role in more accurate road recognition and intelligent functions such as obstacle avoidance.

[0051] The above is a preferred embodiment of the present application. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for defogging an electronic rearview mirror image based on image information integration, characterized in that: Including: S101: Obtain real-time ambient light data and dynamically adjust the contrast and brightness of the image according to the data; S102: Generate an optimized base image based on the adjusted image parameters; S103: Adjust the parameters of the image edge enhancement algorithm according to real-time weather conditions; S104: Apply the adjusted parameters to the base image to output a clear defogged image.

2. The electronic rearview mirror image defogging method based on image information integration according to claim 1, characterized in that: The step S101 includes: Determine the initial brightness enhancement coefficient A0 based on the vehicle speed and the camera viewing angle range. The faster the vehicle speed and the larger the camera viewing angle range, the larger the value of A0; According to the real-time ambient light intensity E, use the brightness adjustment formula: A = A0 * (1 + tanh(E / (Es * Ec))); Where Es is the ambient light intensity threshold, Ec is the ambient light intensity center offset, and tanh is the hyperbolic tangent function, which is used to map the input value to the interval (-1, 1); Dynamically update the photosensitivity of the camera sensor G = Gbase + A * ΔG; Where Gbase is the base sensitivity and ΔG is the photosensitivity gain change; If the current ambient light intensity E is greater than the preset maximum light intensity Em, set A = 1 and keep G unchanged.

3. The electronic rearview mirror image defogging method based on image information integration according to claim 2, characterized in that: The step S102 includes: Based on the precipitation R in the weather information, initially select a suitable fog segmentation strategy S0; After gray processing the image, use the Sobel operator to calculate the gradient. If the average gradient M < Th, use high-pass filtering to enhance the image; Calculate the local contrast enhancement factor Cy: Cy = exp(α * M^2 / σs^2); Where α and σs are the contrast influence coefficient and standard deviation respectively, and exp is the exponential function; Fuse the original image and the filtered image and perform weighted average by pixel: W = (Img_original + C * Img_filter) / 2.

4. The electronic rearview mirror image defogging method based on image information integration according to claim 3, characterized in that: The step S103 includes: Adjust the sharpening intensity K0 = V / Vmax based on the vehicle speed V according to the preset rule. The faster the vehicle speed, the greater the sharpening intensity; Pre-adjust the contrast parameter of the image for the predicted humidity H in the next few frames, and use the following formula for correction: B = β(H - 50), where β is the humidity change response factor; Real-time detect and feedback the current visual visibility VD, and set the visibility compensation coefficient γ(VD) = 7 + 0.1 * sqrt(abs(log(VD / 10))). When VD reaches the set threshold Tvd, reduce the K weight.

5. The electronic rearview mirror image defogging method based on image information integration according to claim 4, characterized in that: The step S104 includes: Obtain the average light intensity Ep_avg and standard deviation σEp within the previous second; When it is detected that the ambient light fluctuates rapidly, that is, abs(Ep_current - Ep_previous) > τ * σEp, where τ is the sensitivity threshold, perform transient adaptive adjustment; Perform color balance optimization ColorAdj, and use the nonlinear mapping relationship of the Lab color space to separate and transform L, a*, b*; Use the formula F(C) = θ * C² + (1 - θ) * C to adjust the interaction between the channels of the color space, where C represents the intensity of any single color channel and θ is the balance adjustment parameter.

6. The electronic rearview mirror image defogging method based on image information integration according to claim 5, characterized in that: The step S103 includes: Precisely select the image edge enhancement degree e0 according to the current atmospheric humidity HR; When HR < hr_limit, execute the low humidity mode: refine the target edge through a bilateral filter; If high humidity triggers the high humidity protection mechanism: F_HighHR = w_α * max(exp(μt * t), w_ω * cos(π * f * d)), t ∈ [t1, t2]; Apply an improved bilateral filtering function to process the image I = F(I, f, d), making the vehicle edges in the fog environment more clearly visible. f is the spatial filtering radius, d is the spatial distance, μt, wt control the smoothing effect in the time dimension, w_ω controls the frequency weight, and hr_limit represents the critical humidity limit.

7. The electronic rearview mirror image defogging method based on image information integration according to claim 6, characterized in that: The step S103 includes: Introduce an intelligent environment evaluation model to predict possible future environmental changes and make adaptive configuration changes in advance; Define a comprehensive evaluation index Zeta to judge the effectiveness of image processing: Zeta(i) = Σ[Wi * log(δi * (Cy - Di))^(η)], Cy is the statistical attribute value of the processed image such as the mean square error, and Di is the corresponding ideal value; Use the maximum Zeta value obtained under specific conditions as the guiding principle to determine the direction of the image defogging strategy in the next cycle; If the calculation results show a significant decrease for two or more consecutive times, start the reset learning stage to re-learn the optimal parameter sets Wi, η.

8. The electronic rearview mirror image defogging method based on image information integration according to claim 7, characterized in that: The step S101 includes: Comprehensively consider the changes in the illumination angle and direction to automatically correct the influence degree factor_dirction of the direct light source. The influence of the direct light source on the image can be calculated according to the illumination angle and direction by establishing an illumination model and corresponding corrections can be made; Divide the collected data into two parts, one part of the training samples is used to construct a short-term trend prediction engine model_shortTermPred, and the other part is the verification data testdata_valiation; Use cross-validation to determine the optimal hyperparameters sets_optimParam of the model, such as learningRate, nHiddenLayers; Enable the special mode when encountering extreme climate phenomena, that is, if WeatherExt == True: useExtremeProcessingAlgo().

9. The electronic rearview mirror image defogging method based on image information integration according to claim 8, characterized in that: The step S104 includes: Establish a personalized memory bank Memory_bank in combination with historical driving scenarios so that users can retrieve previous successful cases for reference and comparison every time they encounter a similar situation; Set the corresponding priority prio according to the terrain class terrainClass where the vehicle is located to arrange subsequent steps; Calculate the number of effective elements N_effective in each frame of the image, and perform additional magnification or enhancement for cases with less than threshold_num. Context-related information contextInfo is introduced to assist in identifying the background noise interference source ContextFilter(contextInfo)=contextImportanceWeightedMean(contextStrength).

10. The electronic rearview mirror image defogging method based on image information integration according to claim 9, characterized in that: The step S103 includes: With the help of deep learning resources provided by the cloud data center, the denoising performance of the local port is continuously improved (NoiseRedutionPerformance_improvememt(CloudResrouce), and the network connection quality networkStableLevel is monitored to ensure stable transmission efficiency; when packet loss or unsatisfactory delay occurs (NetworkQuality<=low_threshold), the emergency offline operation plan ContingencyPlanofflineMode() is started. The denoising performance improvement here can denoise the local image through the deep learning model in the cloud, and then the processing results are fed back to the local port; Check whether the latest version of the software has any unfixed vulnerabilities BugReport_latest, and push the patch package PatchDeployment() in time; Further analyze the driver's habitual actions and habitual preferences habit_patterns to select an appropriate image presentation mode Presentation_mode (habit_pattern); Regularly maintain an online knowledge graph KnowledgeGraph_online to collect new information about road conditions NewRoadInfo and update it in a timely manner; Determine whether to activate certain features feature_activation_status according to the latest traffic regulations TrafficRegulations and real-time weather conditions RealTimeWeaherSituation; In response to emergency braking or other emergencies, EmergencyResponse() quickly captures and saves the accident scene for later review InCydentReviewProcess().

Images