A vehicle headlamp cover dirty state intelligent self-adaptive perception method

Abstract

The application discloses a kind of vehicle headlamp cover dirty state intelligent adaptive perception method, belong to vehicle lamp detection technical field.It includes the following steps: step S1, through gray level camera in the extinguishing state captures a frame lamp cover surface original image;Step S2, the original image of lamp cover surface collected is preprocessed;Step S3, the image after pre-processing is carried out multi-dimensional state feature phasor extraction, and the current characteristic phasor Ft is obtained;Step S4, establish clean state dynamic reference library and carry out online update;Step S5, obtain feature difference vector ΔF.The application provides a kind of vehicle headlamp cover dirty state intelligent adaptive perception method, can output a real-time stable feature difference signal, the current lamp cover surface state is accurately characterized between its self-learning dynamic clean reference by feature difference signal Multi-dimensional deviation, provide high confidence input source for subsequent cleaning decision and control.

Description

Technical Field

[0001] This invention relates to an intelligent adaptive sensing method for the dirt status of vehicle headlight covers, belonging to the field of vehicle headlight detection technology. Background Technology

[0002] Currently, with the deep integration of automotive intelligence and electrification, advanced driver assistance systems and autonomous driving technologies place extremely high demands on the reliability of vehicle perception and execution components. As a crucial active safety and signal indication component, vehicle lights directly impact driving safety at night and in adverse weather conditions.

[0003] Intelligent lighting technologies such as adaptive high beams and projection headlights have become industry trends, and their realization depends on a clear and unobstructed light path. Therefore, developing an "intelligent self-closed-loop vehicle lighting system" capable of accurately sensing the cleanliness of the lampshade in real time and automatically triggering cleaning actions is crucial. The primary technical bottleneck of this system lies in finding a stable, economical, and reliable dirt detection and sensing solution.

[0004] In a vehicle headlight dirt detection and sensing system, one of the core tasks is to accurately detect any changes in the surface condition of the headlight cover and transmit this signal to the evaluation and decision-making module. Whether the sensing scheme is image-based or sensor-based, the basic principle is to compare the current detected signal with a baseline signal representing an ideal clean state. Therefore, a stable, reliable, and long-term trustworthy cleanliness baseline is the foundation for the entire sensing system. However, in actual use, headlight covers undergo slow changes due to ultraviolet radiation, high temperatures, chemical corrosion, and physical wear, resulting in yellowing, minor scratches, and other optical properties that are not static. Therefore, using a fixed cleanliness baseline cannot adapt to the slow changes in headlight cover aging and scratches, leading to decreased reliability and signal distortion in the sensing system over long-term use. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide an intelligent adaptive perception method for the dirt status of vehicle headlight covers. This method can output a real-time stable feature difference signal, which accurately characterizes the multidimensional deviation between the current surface state of the headlight cover and its self-learned dynamic cleanliness benchmark, providing a high-confidence input source for subsequent cleaning decisions and control.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: A method for intelligent adaptive perception of the dirt status of vehicle headlight covers includes the following steps: Step S1: Capture a frame of the original image of the lampshade surface using a grayscale camera in the dark state; Step S2: Preprocess the original image of the lampshade surface that has been acquired; Step S3: Extract multi-dimensional state feature phasors from the preprocessed image to obtain the current feature phasor F. t ; Step S4: Establish a dynamic cleanroom status baseline library and update it online; Step S5: Transfer the current feature vector F t The dynamic benchmark feature vector F is compared with the current feature vector provided by the cleanroom dynamic benchmark library. ref Perform vector subtraction to obtain the feature difference vector ΔF.

[0007] Furthermore, in step S2, the acquired original image of the lampshade surface is preprocessed; specifically, this includes the following steps: Step S21: Define a fixed ROI region in the original image of the lampshade surface; Step S22: Adaptive histogram equalization is used to enhance the contrast of the image within the ROI region; Step S23: Remove random noise from the image within the ROI region using the non-local means denoising method.

[0008] Furthermore, in step S3, multi-dimensional state feature phasors are extracted from the preprocessed image to obtain the current feature phasor F. t Specifically, the steps include the following: Step S31: Calculate the mean μ and standard deviation σ of the gray values ​​of all pixels within the ROI region; Step S32: Calculate the contrast C and homogeneity H of the image within the ROI region under the specified spatial relationship; Step S33: Calculate the mean gradient magnitude G m and gradient magnitude variance G v ; Step S34: Convert the spatial domain image to the frequency domain using Fast Fourier Transform. Divide the spectrum into a low-frequency central region and a high-frequency peripheral region based on a preset radius threshold. Then, based on the low-frequency energy E... low and high-frequency energy E high Calculate the high-frequency energy ratio F h .

[0009] Furthermore, the formula for calculating the mean μ of all pixel grayscale values ​​within the ROI region in step S31 is as follows: ; The formula for calculating the standard deviation σ of all pixel grayscale values ​​within the ROI region is as follows: ; Where N is the total number of pixels within the ROI region; Let be the grayscale value of the i-th pixel within the ROI region.

[0010] Furthermore, the formula for calculating the contrast ratio C in step S32 is as follows: ; The formula for calculating the homogeneity H is as follows: ; Where i and j are gray level index variables; L is the number of gray levels of the image within the ROI region; P(i,j) is the element value of the normalized gray-level co-occurrence matrix at position (i,j).

[0011] Furthermore, in step S33, the average gradient magnitude G m The calculation formula is as follows:

[0012] The gradient magnitude variance G v The calculation formula is as follows:

[0013] Where N is the total number of pixels in the image within the ROI region; This represents the value of the k-th pixel in the gradient magnitude map.

[0014] Furthermore, in step S34, the high-frequency energy ratio F h The calculation formula is as follows:

[0015] Among them, E low For the low-frequency portion of energy, E high This refers to the high-frequency energy component.

[0016] Furthermore, in step S4, a dynamic baseline library of cleanroom conditions is established and updated online; specifically, this includes the following steps: Step S41: When the clean baseline learning conditions are met, the baseline learning mode is entered, M frames of images are continuously acquired, the current feature vector of each frame is extracted, and then the dynamic baseline feature vector F of these M current feature vectors is calculated. ref and dynamic baseline standard deviation vector σ ref , where M is an integer greater than 1; Step S42: After each complete automatic headlight cleaning operation, the current feature vector F is used. t Update the dynamic baseline feature vector F ref and dynamic baseline standard deviation vector σ ref The updated dynamic baseline feature vector is obtained. and the updated dynamic benchmark variance vector .

[0017] Furthermore, the cleanliness benchmark learning condition is: the vehicle has been driving continuously for more than 30 minutes and during this period the vehicle has not triggered any automatic headlight cleaning operation.

[0018] Furthermore, the formula for calculating the feature difference vector ΔF in step S5 is as follows:

[0019] in, F is the current feature vector; ref The dynamic baseline feature vector provided for the current feature vector of the clean state dynamic baseline library.

[0020] By employing the above technical solution, this invention utilizes a dynamic benchmark learning and update mechanism to automatically update the cleanliness benchmark upon which perception relies, tracking changes such as aging and scratches on the headlight cover. This solves the long-standing problem of false alarms and missed alarms caused by fixed benchmark drift, ensuring the accuracy of perception. The feature difference vector output by this invention can maximally isolate the influence of device-specific changes, more realistically reflecting the changes in headlight cover state caused by actual contaminants, providing stable and reliable high-quality input for subsequent advanced decision-making processes such as dirt quantification and type identification. Attached Figure Description

[0021] Figure 1 This is a flowchart of the intelligent adaptive sensing method for the dirty state of vehicle headlight covers according to the present invention. Detailed Implementation

[0022] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0023] like Figure 1 As shown, this embodiment provides an intelligent adaptive sensing method for the dirty state of vehicle headlight covers, including the following steps: Step S1: Capture a frame of the original image of the lampshade surface in the off state using a grayscale camera installed inside the headlight. This avoids strong interference from the headlight's own illumination on the imaging, and obtains a clear image of the lampshade surface with ambient light as the main light source. This results in a single frame of the original image of the lampshade surface, which mainly contains the surface texture and contaminants.

[0024] Step S2: Preprocess the acquired raw image of the lampshade surface. This step aims to improve the image quality of the raw image of the lampshade surface, preparing for feature extraction. The following operations are performed sequentially on the acquired raw image of the lampshade surface: Step S21: Define a fixed ROI region in the original image of the lampshade surface. The ROI should cover the most easily soiled and representative parts of the lampshade, exclude the interference of the lamp body frame, and ensure that all subsequent processing is based on this stable region.

[0025] Step S22: Adaptive histogram equalization technology is used to enhance the contrast of the image within the ROI region, thereby reducing the lighting shadows caused by uneven ambient light and making the texture and details more prominent.

[0026] Step S23: Remove random noise from the image within the ROI region using nonlocal mean denoising, while preserving image edge and texture information and avoiding detail blurring caused by traditional filtering methods.

[0027] Step S3: Extract multi-dimensional state feature phasors from the preprocessed image to obtain the current feature phasor F. t This step converts the image into a measurable feature vector, thereby providing a comprehensive digital description of the current state of the lampshade surface. The current feature vector in this embodiment includes four main categories with a total of seven feature components: global statistical features, texture features, structural features, and frequency domain features. Global statistical features include two feature components: the mean μ and standard deviation σ of the grayscale values. Texture features include two feature components: contrast C and homogeneity H. Structural features include the mean gradient magnitude G. m and gradient magnitude variance G v Two characteristic components, the frequency domain feature includes the high-frequency energy ratio F h A feature component. Specifically: Step S31: Calculate the mean μ and standard deviation σ of all pixel grayscale values ​​within the ROI region. These two features are sensitive to variations in uniformity. The mean μ represents the arithmetic mean of all pixel grayscale values ​​within the image detection and perception region, used to measure the overall brightness level of the region. The standard deviation σ represents the dispersion of each pixel grayscale value within the image detection and perception region relative to the mean μ. A larger standard deviation indicates higher contrast within the region, and potentially richer texture or detail.

[0028] The formula for calculating the mean μ of all pixel grayscale values ​​within the ROI region is as follows: ; The formula for calculating the standard deviation σ of all pixel grayscale values ​​within the ROI region is as follows: ; Where N is the total number of pixels within the ROI region, and is a positive integer. It is determined by the size of the defined fixed ROI region; for example, if the ROI is 100 pixels × 100 pixels, then N = 10000.

[0029] is the gray value of the i-th pixel within the ROI region. In a grayscale image, it is usually an integer between 0 and 255, with black having a gray value of 0 and white having a gray value of 255.

[0030] ∑ is the summation operator, which calculates the summation of a sequence. ~ Perform summation.

[0031] Step S32: Calculate the contrast C and homogeneity H of the image within the ROI region under a specified spatial relationship, for example, with a direction of 0 degrees and a distance of 1 pixel. Contrast C reflects the drastic degree of local gray-level changes in the image, measuring the clarity of texture and the depth of grooves. The larger the value, the deeper the grooves and the more obvious the gray-level differences in local areas of the image. Homogeneity H reflects the uniformity of the image texture and the regularity of the local structure. The larger the value, the more uniform the image texture and the more concentrated the gray-level distribution is on the diagonal, meaning that pixel pairs have similar gray-level values.

[0032] The formula for calculating contrast ratio C is as follows: ; The formula for calculating homogeneity H is as follows: ; Where i and j are gray level index variables, which are integers in the range [0, L-1], representing the gray level of the first and second pixels in a pixel pair, respectively.

[0033] L is the number of gray levels in the image within the ROI region. It is a positive integer that measures the pixel gray level to L levels. For example, for an 8-bit grayscale image, L=256.

[0034] P(i,j) is the element value at position (i,j) of the normalized gray-level co-occurrence matrix. It represents the joint probability of one pixel having gray level i and another pixel having gray level j among all pixel pairs satisfying a specific spatial relationship in an image. First, based on the defined spatial relationship, such as along a 0-degree direction with a 1-pixel interval, the frequency of each gray-level combination (i,j) is counted, forming an L×L matrix. Then, each element in this matrix is ​​divided by the total number of pixel pairs and normalized, ultimately yielding the gray-level co-occurrence matrix P.

[0035] ∑∑ is a double summation operator, which means that the results of the expression under all combinations are summed by iterating through variables i and j from 0 to L-1 respectively.

[0036] Step S33: Dirt significantly alters the edge information of the lampshade surface. For example, a uniform water film and oil film smooth the lampshade surface, resulting in a weaker overall edge and a more uniform distribution, i.e., the average gradient amplitude G. m Decrease and gradient magnitude variance Gv The gradient magnitude is reduced; however, discrete mud dots introduce many strong local edges, leading to potentially enhanced and highly unevenly distributed overall edges, i.e., the average gradient magnitude G. m Increase and gradient magnitude variance G v The gradient magnitude increases. Therefore, the mean gradient magnitude G m and gradient magnitude variance G v It constitutes the current feature vector F t A key component, used for subsequent difference calculations and contamination pattern identification. Gradient magnitude mean G m This reflects the overall edge strength or texture sharpness of the image; a larger value indicates stronger edges and clearer textures in the image. Gradient magnitude variance G v It measures the non-uniformity of gradient intensity distribution in an image. The larger the value, the more uneven the edge intensity distribution, and there may be strong edge regions and weak edge / flat regions at the same time; the smaller the value, the more uniform the edge intensity distribution.

[0037] This embodiment introduces a gradient magnitude map, and the gradient magnitude calculation formula is as follows:

[0038] in, The gradient magnitude map is a two-dimensional matrix. The value of each pixel represents the total gradient intensity at that point. Gradient intensity, also known as edge intensity, is obtained by the geometric synthesis of horizontal and vertical gradients and reflects the degree of grayscale change in the image at that point. x and G y Let be the gradient matrices in the X and Y directions.

[0039] Calculate the mean gradient magnitude G m Gradient magnitude mean G m The calculation formula is as follows:

[0040] Calculate the gradient magnitude variance G v Gradient magnitude variance G v The calculation formula is as follows:

[0041] Where N is the total number of pixels in the image within the ROI region, which is a positive integer; This represents the value of the k-th pixel in the gradient magnitude map.

[0042] Step S34: Convert the spatial domain image to the frequency domain using Fast Fourier Transform. Divide the spectrum into a low-frequency central region and a high-frequency peripheral region based on a preset radius threshold. Then, based on the low-frequency energy E... low and high-frequency energy E highCalculate the high-frequency energy ratio F h High-frequency energy ratio F h This indicates the proportion of high-frequency components in the total energy of an image. F h A higher F value indicates that the image contains richer high-frequency information such as details, edges, or noise; h The lower the value, the smoother the image overall, and the more dominant the low-frequency information. High-frequency energy ratio F h The calculation formula is as follows:

[0043] Among them, the high-frequency part energy E high The low-frequency energy E is obtained by summing the squares of the amplitudes of all complex coefficients within a specified high-frequency region (such as the annular or rectangular region surrounding the center of the spectrum) on the frequency domain graph after performing a two-dimensional discrete Fourier transform on the image. low The low-frequency components, such as the contour and background of the image, are obtained by summing the squares of the amplitudes of all complex coefficients in the low-frequency region (near the center of the image spectrum) after Fourier transform.

[0044] This represents the total energy of the image within the selected frequency band, such that F h It becomes a proportional value relative to the total energy of the image, thereby eliminating the impact of overall changes in absolute brightness or contrast of the image, and the feature is more robust.

[0045] Step S4: Establish a dynamic cleanroom status baseline library and update it online. Specifically: Step S41: The cleanliness status dynamic benchmark library in this embodiment is not static, but a learnable dynamic model. After the vehicle leaves the factory or the system is reset, when the system determines that the cleanliness benchmark learning conditions are met, the cleanliness benchmark learning conditions are: the vehicle has been driving continuously for more than 30 minutes and during this period the vehicle has not triggered any automatic headlight cleaning operation, then the system automatically enters the benchmark learning mode. In this mode, M frames of images are continuously acquired, for example, 20 frames, the current feature vector of each frame is extracted, and then the dynamic benchmark feature vector F of these 20 current feature vectors is calculated. ref and dynamic baseline standard deviation vector σ ref .

[0046] Dynamic baseline eigenvector F ref It is a seven-dimensional column vector, representing the seven components in step S3. This vector serves as a reference benchmark for assessing whether the current state is dirty. The dynamic benchmark eigenvector F of the current eigenvector. ref The calculation formula is as follows:

[0047] in, The baseline grayscale mean represents the long-term average of pixel grayscale values ​​in the cleanroom lampshade image within the ROI region, which is the historical estimate of the global statistical feature mean μ.

[0048] The baseline grayscale standard deviation represents a long-term estimate of the range of pixel grayscale value fluctuations in the cleanroom lampshade image within the ROI region, which is also the historical estimate of the global statistical characteristic standard deviation σ.

[0049] The baseline contrast is a long-term estimate of the texture contrast characteristics of the cleanroom lampshade image, which is the historical estimate of the contrast C.

[0050] The baseline homogeneity represents the long-term estimate of the homogeneity characteristics of the cleanroom lampshade image texture, which is the historical estimate of homogeneity H.

[0051] The mean gradient magnitude, representing a long-term estimate of the overall edge intensity characteristics of the cleanroom lampshade image, is the mean gradient magnitude G. m Historical estimates.

[0052] The baseline gradient magnitude variance, representing a long-term estimate of the uniformity of edge intensity distribution in the cleanroom lampshade image, is denoted as G. v Historical estimates.

[0053] The high-frequency energy ratio, representing a long-term estimate of the frequency domain energy distribution characteristics (high-frequency proportion) of the cleanroom lamp cover image, is the baseline high-frequency energy ratio F. h Historical estimates.

[0054] Dynamic benchmark standard deviation vector σ ref It is a seven-dimensional column vector, where each component stores the corresponding dynamic baseline feature vector F in the clean state. ref An estimate of the normal fluctuation range caused by inherent factors such as minor fluctuations in ambient light and image noise is used to normalize each feature when calculating the degree of difference. Dynamic baseline standard deviation vector σ ref The calculation formula is as follows:

[0055] in, for The estimated standard deviation under clean conditions. for The estimated standard deviation under clean conditions. for The estimated standard deviation under clean conditions. for The estimated standard deviation under clean conditions. for The estimated standard deviation under clean conditions. for The estimated standard deviation under clean conditions. for The estimated standard deviation under clean conditions. For example, It is calculated from multiple samples. The standard deviation represents the inherent fluctuation range of the overall brightness of an image under clean conditions.

[0056] Step S42: In daily use, after the vehicle completes a full automatic headlight cleaning operation, the current feature vector F is used... t Update the dynamic baseline feature vector F ref and dynamic baseline standard deviation vector σ ref The updated dynamic baseline feature vector is obtained. and the updated dynamic benchmark variance vector .

[0057] Updated dynamic baseline feature vector The calculation formula is as follows:

[0058] in, This is the dynamic baseline feature vector before the update.

[0059] F t This is the current feature vector, which is the current feature vector extracted in step S3.

[0060] α is the forgetting factor, a pre-defined constant with a value range of 0 < α < 1, typically close to 1, such as 0.95. α controls the weight of historical information and new observations in the update process. The larger the α value, the more conservative the update, the higher the proportion of historical benchmarks, and the slower the system tracks changes in the lampshade, but the less sensitive it is to transient dirt.

[0061] Updated dynamic benchmark variance vector The calculation formula is as follows:

[0062] in, Each component represents the latest estimate of the fluctuation range of the corresponding characteristic value under clean conditions.

[0063] This is the dynamic baseline variance vector before the update.

[0064] F tThis is the current feature vector.

[0065] α is the forgetting factor.

[0066] It is used to measure the degree of dispersion of current observations relative to a new benchmark.

[0067] Step S5: Transfer the current feature vector F t The dynamic benchmark feature vector F is compared with the current feature vector provided by the cleanroom dynamic benchmark library. ref Performing vector subtraction yields the feature difference vector ΔF, where each component represents the specific difference between the current lampshade state and the cleanliness baseline in a particular image feature dimension. The formula for calculating the feature difference vector ΔF is as follows: .

[0068] Finally, the calculated feature difference vector ΔF is used as the output of this method, which is provided to the next-level decision system as the basis for judging whether the vehicle is dirty and the degree of dirtiness. The next-level decision system can make a judgment based on the feature difference vector ΔF, generate a cleaning command, and drive the vehicle's headlight automatic cleaning actuator to perform physical cleaning actions.

[0069] The specific embodiments described above further illustrate the technical problems, technical solutions, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent adaptive perception of the dirt status of vehicle headlight covers, characterized in that, Includes the following steps: Step S1: Capture a frame of the original image of the lampshade surface using a grayscale camera in the dark state; Step S2: Preprocess the original image of the lampshade surface that has been acquired; Step S3: Extract multi-dimensional state feature phasors from the preprocessed image to obtain the current feature phasor F. t ; Step S4: Establish a dynamic cleanroom status baseline library and update it online; Step S5: Transfer the current feature vector F t The dynamic benchmark feature vector F is compared with the current feature vector provided by the cleanroom dynamic benchmark library. ref Perform vector subtraction to obtain the feature difference vector ΔF.

2. The intelligent adaptive sensing method for the dirty state of vehicle headlight covers according to claim 1, characterized in that, In step S2, the original image of the lampshade surface acquired is preprocessed; specifically, the following steps are included: Step S21: Define a fixed ROI region in the original image of the lampshade surface; Step S22: Adaptive histogram equalization is used to enhance the contrast of the image within the ROI region; Step S23: Remove random noise from the image within the ROI region using the non-local means denoising method.

3. The intelligent adaptive sensing method for the dirty state of vehicle headlight covers according to claim 1, characterized in that, In step S3, multi-dimensional state feature phasors are extracted from the preprocessed image to obtain the current feature phasor F. t Specifically, the steps include the following: Step S31: Calculate the mean μ and standard deviation σ of the gray values ​​of all pixels within the ROI region; Step S32: Calculate the contrast C and homogeneity H of the image within the ROI region under the specified spatial relationship; Step S33: Calculate the mean gradient magnitude G m and gradient magnitude variance G v ; Step S34: Convert the spatial domain image to the frequency domain using Fast Fourier Transform. Divide the spectrum into a low-frequency central region and a high-frequency peripheral region based on a preset radius threshold. Then, based on the low-frequency energy E... low and high-frequency energy E high Calculate the high-frequency energy ratio F h .

4. The intelligent adaptive sensing method for the dirty state of vehicle headlight covers according to claim 3, characterized in that, The formula for calculating the mean μ of all pixel grayscale values ​​within the ROI region in step S31 is as follows: ； The formula for calculating the standard deviation σ of all pixel grayscale values ​​within the ROI region is as follows: ； Where N is the total number of pixels within the ROI region; Let be the grayscale value of the i-th pixel within the ROI region.

5. The intelligent adaptive sensing method for the dirty state of vehicle headlight covers according to claim 3, characterized in that, The formula for calculating the contrast ratio C in step S32 is as follows: ； The formula for calculating the homogeneity H is as follows: ； Where i and j are gray level index variables; L is the number of gray levels of the image within the ROI region; P(i,j) is the element value of the normalized gray-level co-occurrence matrix at position (i,j).

6. The intelligent adaptive sensing method for the dirty state of vehicle headlight covers according to claim 3, characterized in that, In step S33, the average gradient magnitude G m The calculation formula is as follows: ； The gradient magnitude variance G v The calculation formula is as follows: ； Where N is the total number of pixels in the image within the ROI region; This represents the value of the k-th pixel in the gradient magnitude map.

7. The intelligent adaptive sensing method for the dirty state of vehicle headlight covers according to claim 3, characterized in that, In step S34, the high-frequency energy ratio F h The calculation formula is as follows: ； Among them, E low For the low-frequency portion of energy, E high This refers to the high-frequency energy component.

8. The intelligent adaptive sensing method for the dirt status of vehicle headlight covers according to claim 1, characterized in that, In step S4, a dynamic baseline library of cleanroom conditions is established and updated online; specifically, this includes the following steps: Step S41: When the clean baseline learning conditions are met, the baseline learning mode is entered, M frames of images are continuously acquired, the current feature vector of each frame is extracted, and then the dynamic baseline feature vector F of these M current feature vectors is calculated. ref and dynamic baseline standard deviation vector σ ref , where M is an integer greater than 1; Step S42: After each complete automatic headlight cleaning operation, the current feature vector F is used. t Update the dynamic baseline feature vector F ref and dynamic baseline standard deviation vector σ ref The updated dynamic baseline feature vector is obtained. and the updated dynamic benchmark variance vector .

9. The intelligent adaptive sensing method for the dirty state of vehicle headlight covers according to claim 8, characterized in that, The cleanliness baseline learning condition is: the vehicle has been driven continuously for more than 30 minutes and no automatic headlight cleaning operation has been triggered during this period.

10. The intelligent adaptive sensing method for the dirty state of vehicle headlight covers according to claim 1, characterized in that, The formula for calculating the feature difference vector ΔF in step S5 is as follows: ； in, F is the current feature vector; ref The dynamic baseline feature vector provided for the current feature vector of the clean state dynamic baseline library.

Images