Electric swap station vehicle bottom monitoring method based on AI vision

Abstract

The invention relates to the technical field of image analysis, in particular to a method for monitoring the bottom of a battery swap station vehicle based on AI vision, which comprises the following steps: acquiring an image frame to extract bottom gray level distribution, calculating illumination difference to judge an abnormal frame, extracting a gray level jump band, calculating a slope, screening miscellaneous points, identifying a closed contour, and extracting a texture direction and a pitch. And analyzing an offset screening abnormal number, calculating an included angle and a contraction mark mutation pixel, and generating a bottom monitoring result. According to the method, the image stability under the complex illumination condition is enhanced by analyzing the gray level distribution and illumination change in the image frame, the edge response density difference elimination mechanism is combined, the structure overlapping interference is avoided, the closure degree of the closed structure and the texture period are extracted, and the structural integrity recognition is enhanced; texture direction and pitch cross-frame change analysis supports micro structure offset early warning, boundary point included angles are combined with contour shrinkage trends, accurate marking of abnormal areas is achieved, and sensitivity, accuracy and reliability of bottom monitoring are improved.

Description

Technical Field

[0001] The present invention relates to the field of image analysis technology, and in particular to a method for monitoring the underside of vehicles at a battery swap station based on AI vision. Background Art

[0002] The field of image analysis technology encompasses image processing, analysis, and understanding, and is widely used in fields such as computer vision, autonomous driving, and medical imaging. The core of this technology lies in processing and analyzing images or videos to extract useful information or provide decision support. Image analysis technology involves multiple steps, including image acquisition, preprocessing, feature extraction, and pattern recognition, encompassing a variety of algorithms and methods, such as edge detection, object detection, image segmentation, image matching, and deep learning. Technologies in this field are widely used in scenarios such as surveillance, robotics, automated manufacturing, and virtual reality, and are an indispensable component of intelligent devices and systems.

[0003] Among them, the AI vision-based vehicle bottom monitoring method at battery swap stations refers to the use of artificial intelligence vision technology to monitor the bottom of electric vehicles at battery swap stations. This method mainly involves processing and analyzing real-time images or video data of the bottom of the vehicle, and using computer vision technology to detect and identify the bottom structure, components, damage, etc. This method uses AI vision technology to achieve automated monitoring, does not rely on manual inspection, and improves the accuracy and efficiency of monitoring. Specifically, this patented technology includes steps such as vehicle bottom image acquisition, image preprocessing, feature extraction and target recognition, relying on image analysis methods to achieve comprehensive monitoring and fault identification of the vehicle bottom.

[0004] In practical applications, existing technologies suffer from insufficient response to changes between consecutive image frames. The lack of cross-frame feature tracking mechanisms results in a sluggish response to subtle structural changes and dynamic deformations, making it difficult to meet the structural stability requirements in high-speed operating areas. Image processing methods often rely on static feature extraction within frames, ignoring the impact of illumination fluctuations on image content recognition boundaries. This can easily lead to recognition errors and regional jumps under sudden illumination changes. Edge detection strategies often rely on single-pixel gradients or morphological operations, lacking a density-based cleaning process. This makes it difficult to eliminate non-target responses in areas with dense grayscale jumps, compromising the reliability of boundary results. At the structural feature level, existing image analysis pathways fail to address complex metrics such as contour closure and texture periodic stability, resulting in incomplete or aliased recognition results for closed structures. Recognition responses are slow for micro-scale anomalies such as loose assembly and structural misalignment, making it difficult for monitoring images to reflect key change nodes. This increases the lag in fault warnings, further impacting the real-time effectiveness of the entire monitoring system and its ability to determine operational safety boundaries. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the existing technology and propose a vehicle bottom monitoring method for battery swap stations based on AI vision.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for monitoring the underside of a vehicle at a battery swap station based on AI vision, comprising the following steps:

[0007] S1: Obtain image frames of the battery swapping operation area, extract the grayscale distribution of the vehicle bottom contour, calculate the illumination difference and photosensitivity mean change between adjacent frames, identify frames with illumination fluctuation amplitude exceeding the mean as abnormal frames, and generate a photosensitivity distribution enhancement layer sequence;

[0008] S2: Based on the photosensitive distribution enhanced layer sequence, extract the grayscale jump band of the battery compartment closed groove and the chute hook angle area, calculate the slope extreme value, filter the pixel mean of the non-edge area, remove the noise in the edge error band, and generate a boundary response cleaned image set;

[0009] S3: calling the boundary response to clean the closed contour pixels in the image set, calculating the boundary closure and pixel continuity, screening the areas with high closure, and generating a closed structure texture direction group;

[0010] S4: calling the structure area in the closed structure texture direction group, obtaining the texture main axis direction and pitch length, calculating the direction difference and pitch offset between adjacent frames, screening the area numbers with excessive offset, and generating a structure texture offset rate list;

[0011] S5: Call the abnormal area number in the structure texture offset rate list, extract the boundary point position, calculate the angle change and contour shrinkage, mark the pixels with the direction mutation angle greater than the threshold, and generate the vehicle bottom monitoring result.

[0012] As a further solution of the present invention, the photosensitivity distribution enhancement layer sequence includes photosensitivity abnormality frame index, grayscale jump distribution layer, and regional illumination change feature label; the boundary response cleaning image set includes edge response layer, non-edge area mean map, and boundary pixel difference filtering result; the closed structure texture direction group includes high closure structure contour, main axis texture direction classification, and repeated texture pitch feature; the structure texture offset rate list includes texture direction change amplitude, pitch offset ratio parameter, and abnormal structure number index; the vehicle bottom monitoring result includes boundary angle change distribution, contour shrinkage rate index, and direction mutation area mark.

[0013] As a further solution of the present invention, the specific steps of S1 are:

[0014] S101: Obtain a sequence of image frames captured by an image acquisition device in the slide-in section of the battery swapping operation area, extract the grayscale channel pixel distribution of the vehicle bottom contour area in each frame of the image, calculate the light intensity difference based on the grayscale mean and distribution difference in the same area of adjacent image frames, and analyze the sensitivity change rate to obtain the grayscale difference and sensitivity change rate range;

[0015] S102: Based on the grayscale difference and the sensitivity change rate interval, the illumination change value of each frame image is numerically compared with the average illumination fluctuation amplitude to determine whether the illumination fluctuation amplitude exceeds the average illumination fluctuation threshold, the corresponding frame image is screened as an abnormal frame, and an index set of sensitivity abnormal frames is obtained;

[0016] S103: Calling the grayscale channel pixel values of the image frames in the photosensitivity abnormality frame index set, detecting the grayscale mutation area in the image, extracting the grayscale jump area by classifying the jump amplitude and distribution morphology, and generating a photosensitivity distribution enhancement layer sequence according to the category mark.

[0017] As a further solution of the present invention, the specific steps of S2 are:

[0018] S201: Based on the image frames in the photosensitive distribution enhancement layer sequence, extract the grayscale channel pixel values of the corner area where the edge segment of the battery compartment closed groove and the chute hook are connected, identify the grayscale jump band, and generate a grayscale jump band distribution layer by calculating the grayscale value changes within the image area;

[0019] S202: Calculate the slope extreme points of the grayscale value sequence based on the grayscale jump band distribution layer, determine the edge response concentration segment in the region, filter the pixel distribution mean of the non-edge region, and eliminate the density difference with the pixel value of the edge response band to obtain the edge response cleaned layer;

[0020] S203: performing a difference elimination operation according to the edge response cleaning layer, processing the density difference between the boundary pixels and the reference value, screening out the boundary area, and generating a boundary response cleaning image set.

[0021] As a further solution of the present invention, the slope extreme point calculation formula of the gray value sequence is specifically:

[0022]

[0023] Among them, S max Represents the slope extreme point of the gray value sequence, P i Represents the grayscale value of the i-th pixel in the image, P i-1 Represents the grayscale value of the i-1th pixel, Δx i represents the pixel interval between the i-th pixel and the previous pixel, n is the total number of pixels in the grayscale value sequence, and |·| represents the absolute value.

[0024] As a further solution of the present invention, the specific steps of S3 are:

[0025] S301: calling the closed contour pixel set in the boundary response cleansing image set, calculating the boundary closure and pixel continuity of each closed structure in the image, screening areas with closure higher than a set ratio, and generating a structure contour unit set;

[0026] S302: Classifying the texture direction angles of the pixel groups within the structure according to the structure contour unit set, calculating the length of the repeated pitch within each structure region, grouping the texture direction groups according to the pitch variation trend, and obtaining texture direction and pitch classification results;

[0027] S303: Based on the texture direction and pitch classification results, statistical processing is performed, and iterative grouping is performed according to the texture main axis angle difference and the pitch change trend to generate a closed structure texture direction group.

[0028] As a further solution of the present invention, the calculation formula for the repeated pitch length in each structural area is specifically:

[0029]

[0030] Among them, L rep Represents the length of the repeated pitch in each structural area, G j Represents the grayscale value of the jth pixel in the image, G j-p represents the grayscale value of the jpth pixel in the image, p is a fixed step size, which indicates the pixel interval considered in the grayscale change process, m is the total number of pixels in the image, and |·| indicates taking the absolute value.

[0031] As a further solution of the present invention, the specific steps of S4 are:

[0032] S401: calling the structure contour marked area in the closed structure texture direction group, obtaining the texture main axis direction angle and pitch length in the differential image frame corresponding to the structure area with the same number, calculating the texture main axis direction difference of the area, analyzing the pitch change amplitude, and obtaining the direction angle and pitch difference value;

[0033] S402: Calculating an offset ratio index of a structural region in adjacent image frames based on the direction angle and the pitch difference value, comparing the index with a preset pitch stability reference value, and selecting the region numbers whose offset ratios are higher than the reference value to obtain the region numbers with abnormal offset ratios;

[0034] S403: According to the number of the region with abnormal offset ratio, the offset amplitude and distribution trend of the corresponding region are extracted to generate a list of structural texture offset rates.

[0035] As a further solution of the present invention, the calculation formula of the offset ratio index of the structure area in adjacent image frames is specifically:

[0036]

[0037] Among them, P offset Represents the offset ratio index of the structure area in adjacent image frames, ΔX i represents the horizontal displacement change of the i-th structural region, ΔT i represents the time interval of the i-th structural region, m is the total number of structural regions, |·| represents the absolute value, and γ is a constant.

[0038] As a further solution of the present invention, the specific steps of S5 are:

[0039] S501: calling the structure number area marked with abnormal offset ratio in the structure texture offset rate list, extracting the boundary point distribution of the corresponding area in the image, calculating the angle change trend between the boundary points, and obtaining the angle change trend value;

[0040] S502: Calculating the closed contour shrinkage rate of the continuous region of boundary points based on the angle change trend value, and screening the continuous pixel segments whose direction mutation angle is greater than the standard threshold by comparing the calculated value with a preset standard threshold, to obtain the region where the direction mutation angle is greater than the threshold;

[0041] S503: Mark the position of the area where the sudden change angle of the direction is greater than the threshold, perform verification, and generate a vehicle bottom monitoring result.

[0042] Compared with the prior art, the advantages and positive effects of the present invention are:

[0043] In the present invention, by analyzing the grayscale distribution and illumination changes in the image frame, the image stability under complex illumination conditions is enhanced, and the edge response density difference elimination mechanism is combined to avoid structural overlap interference, the closure degree and texture period of the closed structure are extracted, and the structural integrity recognition is strengthened. The analysis of texture direction and pitch changes across frames supports early warning of small structure offsets, and the boundary point angle is combined with the contour shrinkage trend to achieve accurate marking of abnormal areas and improve the sensitivity, accuracy and reliability of bottom monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 Schematic diagram of the steps of the present invention;

[0045] Figure 2 is a flow chart of the steps of S1 of the present invention;

[0046] Figure 3 This is a flow chart of the steps of S2 of the present invention;

[0047] Figure 4 This is a flow chart of the steps of S3 of the present invention;

[0048] Figure 5 This is a flow chart of the steps of S4 of the present invention;

[0049] Figure 6 This is a flow chart of the steps of S5 of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0051] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0052] See also Figure 1 , a vehicle bottom monitoring method at a battery swap station based on AI vision, comprising the following steps:

[0053] S1: Obtain a sequence of image frames captured by the image acquisition device in the slide-in section of the battery swap operation area, extract the grayscale channel pixel distribution within the vehicle bottom contour area in each frame, calculate the ambient light intensity difference and the photosensitivity mean change rate of adjacent image frames in the frame sequence, identify image frames with light intensity fluctuations exceeding the average threshold as photosensitivity abnormal frames, match the grayscale jump distribution layers of the photosensitivity areas in the image and classify and mark them, and generate a photosensitivity distribution enhancement layer sequence;

[0054] S2: Based on the image frames in the photosensitive distribution enhancement layer sequence, the grayscale jump band in the edge segment of the battery compartment closed slot and the chute hook connection corner area is extracted. The slope extreme points of the grayscale value sequence in the area are calculated and the edge response concentrated segment is determined. The pixel distribution mean in the non-edge area is selected as the denoising reference value. The density difference between the boundary pixel value and the reference value in the edge response band is subjected to difference elimination operation to generate the boundary response cleaned image set.

[0055] S3: Calling the boundary response to clean the closed contour pixel set in the image set, calculating the closure and pixel continuity of each closed structure boundary in the image, screening the areas with closure higher than the set ratio as structure contour units, performing angle classification and repeated pitch length calculation on the texture direction of the pixel group inside the structure, and performing grouping according to the texture main axis angle difference and pitch change trend to generate a closed structure texture direction group;

[0056] S4: calling the structure contour marked area in the closed structure texture direction group, obtaining the texture main axis direction angle and pitch length of the same numbered structure area in the differential image frame, calculating the direction difference and pitch offset ratio between adjacent frames, determining the area number whose offset ratio is higher than the pitch stability reference value, extracting the offset amplitude and distribution trend according to the number, and generating a structure texture offset rate list;

[0057] S5: Call the structure number area marked with abnormal offset ratio in the structure texture offset rate list, extract the distribution of boundary points of the corresponding area in the image, calculate the angle change trend and closed contour shrinkage rate between boundary points, mark the position of continuous pixel segments with direction mutation angles greater than the standard threshold, and generate vehicle bottom monitoring results.

[0058] The photosensitivity distribution enhancement layer sequence includes photosensitivity abnormality frame index, grayscale jump distribution layer, and regional illumination change feature label. The boundary response cleaning image set includes edge response layer, non-edge area mean map, and boundary pixel difference filtering results. The closed structure texture direction group includes high closure structure contour, main axis texture direction classification, and repeated texture pitch feature. The structure texture offset rate list includes texture direction change amplitude, pitch offset ratio parameter, and abnormal structure number index. The vehicle bottom monitoring results include boundary angle change distribution, contour shrinkage rate index, and direction mutation area mark.

[0059] See also Figure 2 , the specific steps of S1 are:

[0060] S101: Obtain a sequence of image frames captured by an image acquisition device in the slide-in section of the battery swapping operation area, extract the grayscale channel pixel distribution of the vehicle bottom contour area in each frame of the image, calculate the light intensity difference based on the grayscale mean and distribution difference in the same area of adjacent image frames, and analyze the sensitivity change rate to obtain the grayscale difference and sensitivity change rate range;

[0061] First, by acquiring a sequence of image frames captured by an image acquisition device in the slide-in section of the battery swapping operation area, the grayscale channel pixel distribution of the vehicle bottom contour area in the image sequence can be extracted. For the vehicle bottom area in each frame of the image, its grayscale distribution is analyzed frame by frame, and the grayscale value of each pixel in the area is calculated. Then, by comparing the grayscale mean and distribution difference of adjacent image frames in the same area, the light intensity difference is calculated. The specific execution process includes: taking the difference between the grayscale mean of the current image frame and the previous frame. The result reflects the change in light intensity. Afterwards, the light fluctuation amplitude is calculated in combination with the variance of the grayscale distribution to evaluate the sensitivity change rate. Finally, by setting a reasonable grayscale difference and sensitivity change rate range, such as the grayscale difference can be set between 0-50, and the sensitivity change rate can be set between 0.1-0.5, this is used as the light intensity difference evaluation standard to judge the light change between different image frames. During this process, the illumination intensity difference is calculated by sampling the grayscale mean difference of certain image frames. For example, assuming that the grayscale mean of an image frame is 180, and the grayscale mean of the adjacent frame is 170, the grayscale difference is 10. If this value exceeds the set grayscale difference threshold and the sensitivity change rate is also within the set range, it is considered that the illumination change of the frame image is significant and meets the abnormal judgment conditions.

[0062] S102: Based on the grayscale difference and the sensitivity change rate interval, the illumination change value of each frame is compared with the average illumination fluctuation amplitude to determine whether the illumination fluctuation amplitude exceeds the average illumination fluctuation threshold, the corresponding frame image is screened as an abnormal frame, and an index set of abnormal sensitivity frames is obtained;

[0063] First, it is necessary to obtain the illumination change value and average illumination fluctuation amplitude of the image frame. The illumination change value is calculated by the above-mentioned illumination intensity difference, while the average illumination fluctuation amplitude is calculated by performing grayscale value statistics on multiple consecutive frames of images and calculating the average grayscale fluctuation range within a certain time window. For example, if the illumination fluctuation amplitude of 5 consecutive frames of images is less than 20, it is considered normal fluctuation, and if the illumination change value is greater than 20, it is considered abnormal fluctuation. Next, each frame of the image is compared, and the illumination change value is numerically compared with the average illumination fluctuation amplitude. If the illumination change amplitude of a frame of image is greater than the set threshold (such as 20) and its sensitivity change rate is within the preset range (such as 0.2-0.4), the frame of image is marked as an abnormal frame. Through this method, screening is performed to finally form an index set of abnormal sensitivity frames. In this way, the image frame index can be effectively used to calibrate the image frame set with significant illumination fluctuations, providing key information for subsequent image processing.

[0064] S103: Calling the grayscale channel pixel values of the image frames in the sensitivity abnormality frame index set, detecting grayscale mutation areas in the image, extracting the grayscale mutation areas by classifying the jump amplitude and distribution morphology, and generating a sensitivity distribution enhancement layer sequence according to the category label;

[0065] First, the grayscale channel pixel values of each frame of the image are obtained to determine the areas where the grayscale values have changed suddenly. Abnormal areas can be screened by setting a grayscale threshold (for example, the grayscale value change amplitude is greater than 30). Secondly, the jump amplitude of these mutation areas is quantitatively analyzed to determine whether the change amplitude is greater than the set threshold (for example, if the grayscale jump amplitude in a certain area exceeds 30, it is marked as a mutation area). Next, the distribution morphology of the mutation area is analyzed. Image morphological methods such as dilation and erosion operations are used to analyze the morphological characteristics of the mutation area, such as whether the mutation is isolated or concentrated in a local area. Based on these analysis results, the grayscale mutation areas are classified according to different morphologies, and a corresponding photosensitive distribution enhancement layer sequence is generated for each type of grayscale jump area. The enhancement layer displays the visual information of the photosensitive change area to facilitate subsequent analysis.

[0066] See also Figure 3 , the specific steps of S2 are:

[0067] S201: Based on the image frames in the photosensitive distribution enhancement layer sequence, the grayscale channel pixel values of the corner area where the edge of the battery compartment closed groove and the chute hook are connected are extracted, the grayscale jump band is identified, and the grayscale jump band distribution layer is generated by calculating the grayscale value changes within the image area;

[0068] First, based on the image frames in the photosensitive distribution enhancement layer sequence, the grayscale channel pixel values of the battery compartment enclosed groove edge segment and the corner area where the chute hook connects are extracted to obtain the specific lighting characteristics of this area. During execution, each pixel in the image frame contains a grayscale channel value. Extracting these values requires regional segmentation of the battery compartment enclosed groove edge segment and the corner area where the chute hook connects. By cropping a specific area in each image frame, the pixel grayscale distribution of this area is calculated. In a practical application, for example, in the edge area of the battery compartment, the grayscale value of a certain area in this part of the image may jump from 130 to 180. This sudden change represents a change in lighting conditions. Next, by calculating the grayscale value change within this area, a grayscale transition band is formed. In specific implementation, the grayscale value change of each pixel in the area is compared with the grayscale value of the adjacent areas to determine the distribution of the grayscale transition band. If the grayscale value in a certain area suddenly changes, such as from 160 to 250, and the number of pixels in this area reaches a certain proportion, the area is considered to be a grayscale jump band. Finally, the corresponding grayscale jump band distribution layer is generated, and the shape of the grayscale change band is displayed in a visual way.

[0069] S202: Calculate the slope extreme points of the grayscale value sequence based on the grayscale jump band distribution layer, determine the edge response concentration segment in the area, filter the pixel distribution mean of the non-edge area, and eliminate the density difference with the pixel value of the edge response band to obtain the edge response cleaning layer;

[0070] The calculation formula for the slope extreme point of the gray value sequence is as follows:

[0071]

[0072] Among them, S max Represents the slope extreme point of the gray value sequence, P i Represents the grayscale value of the i-th pixel in the image, P i-1 Represents the grayscale value of the i-1th pixel, Δx i represents the pixel interval between the i-th pixel and the previous pixel, n is the total number of pixels in the grayscale value sequence, and |·| represents the absolute value;

[0073] Gray value sequence P i and P i-1 :

[0074] Images are obtained through grayscale processing. Each pixel has a grayscale value, typically ranging from 0 to 255, representing the pixel's brightness. For example, the grayscale value of the first pixel in an image, P1, is 120, and the grayscale value of the second pixel, P2, is 125. Grayscale values are obtained by a camera or image acquisition system and processed and extracted using standard image processing tools such as OpenCV.

[0075] Pixel spacing Δx i :

[0076] Each pixel is spaced Δx i Represents the spatial distance between pixels in an image. For consecutive pixels, the interval between each pixel is assumed to be 1 pixel, meaning the physical distance between each two pixels remains constant. Pixel interval is generally quantified based on image resolution. In digital images, pixels are usually used as the unit. For example, if the image resolution is 1920x1080, the horizontal interval between each pixel is 1 pixel.

[0077] Calculation process:

[0078] Assume that the grayscale values of 5 consecutive pixels in the image are:

[0079] P1=120, P2=125, P3=130, P4=135, P5=140.

[0080] Assume that the pixel interval Δx=1 pixel.

[0081] When calculating the slope extreme value, you first need to calculate the difference in grayscale values:

[0082] P2-P1=125-120=5, P3-P2=130-125=5, P4-P3=135-130=5, P5-P4=140-135=5;

[0083] Then, for each pair of pixels, the grayscale value difference is calculated multiplied by the pixel spacing:

[0084] (P2-P1)·Δx=5·1=5, (P3-P2)·Δx=5·1=5, (P4-P3)·Δx=5·1=5, (P5-P4)·Δx=5·1=5;

[0085] Compute the sum of all differences:

[0086]

[0087] Next, calculate the sum of the squares of the pixel spacing:

[0088]

[0089] Finally, the slope extremes are calculated:

[0090]

[0091] Result interpretation:

[0092] The calculated result S max =10 represents the slope extreme value of the gray value sequence, indicating the degree of change in the gray value of the image. In this example, the smaller slope (5) reflects the relatively stable change of the gray value. The result of the slope extreme value S max =10 illustrates the smoothness and stability of the overall grayscale change, which is of great reference value for subsequent edge response analysis.

[0093] S203: performing a difference elimination operation based on the edge response cleaning layer, processing the density difference between the boundary pixels and the reference value, screening out the boundary area, and generating a boundary response cleaning image set;

[0094] First, for each frame of the image, the boundary area needs to be compared with the set reference value. The reference value can be the overall grayscale mean of the image frame, or the average grayscale value of a specific area. Assuming the reference value is 120 and the grayscale value of the boundary pixel is 140, the density difference between the boundary pixel and the reference value is 140-120=20. Then, the difference is processed, such as eliminating pixels with smaller differences, or eliminating them at a specific threshold. For example, if the threshold is set to 15, any pixel that differs from the reference value by no more than 15 will be eliminated. This operation helps to screen out significant boundary areas and generate a boundary response cleaning image set, which contains processed boundary response areas after removing interference. Ultimately, these boundary areas will serve as key areas for subsequent processing.

[0095] See also Figure 4 , the specific steps of S3 are:

[0096] S301: calling the boundary response to clean the closed contour pixel set in the image set, calculating the boundary closure and pixel continuity of each closed structure in the image, screening the areas with closure higher than a set ratio, and generating a structure contour unit set;

[0097] First, the boundary response cleans the closed contour pixel set in the image set. The boundary closure and pixel continuity calculations are performed for each closed structure in the image. The goal is to assess the integrity and boundary continuity of each structure. This process begins by extracting the closed contour pixel set. These pixels represent edge information in the image and represent the structure's boundary. Next, boundary closure is calculated, which involves determining whether the boundary forms a closed loop. For example, if the closed structure pixels in a region of the image extend along a path and ultimately connect to the starting point to form a closed boundary, the region is considered closed. Incompletely closed regions require further analysis. If the missing portion is small, it may be supplemented using an interpolation algorithm. If the structure's closure exceeds a set percentage (e.g., greater than 95%), the region is considered a valid closed structure. Next, pixel continuity is calculated for the structure, determining whether the edge pixels form a continuous line or are broken. Regions with good continuity, such as edges connected into a smooth line segment, are considered to have high pixel continuity. Through this series of calculations, regions with high closure and good continuity are selected, ultimately generating a set of structural contour units.

[0098] S302: Based on the structure contour unit set, the pixel groups within the structure are classified by angle of texture direction, the length of repeated pitches within each structure region is calculated, the texture direction groups are grouped according to pitch variation trends, and the texture direction and pitch classification results are obtained;

[0099] The calculation formula for the repeated pitch length in each structural area is as follows:

[0100]

[0101] Among them, L rep Represents the length of the repeated pitch in each structural area, G j Represents the grayscale value of the jth pixel in the image, G j-p represents the grayscale value of the jpth pixel in the image, p is a fixed step size, which indicates the pixel interval considered in the grayscale change process, m is the total number of pixels in the image, and |·| indicates taking the absolute value;

[0102] Gray value sequence G j and G j-p :

[0103] Image grayscale values are acquired by image acquisition devices (such as cameras) or image processing algorithms (such as grayscale conversion algorithms). Each pixel's grayscale value represents its brightness and typically ranges from 0 to 255. For example, the grayscale value G1 of the first pixel in an image is 120, while the grayscale value G2 of the second pixel is 125, indicating the brightness variation within that region. Image grayscale values are calculated from the original color image using a grayscale conversion algorithm.

[0104] Pixel spacing p:

[0105] The pixel interval p is the distance between adjacent pixels to calculate the grayscale change, which is usually defined as 1 pixel, that is, the interval between each pixel is 1. The pixel interval in the image is usually fixed, and for most standard digital images, the pixel interval is 1 pixel.

[0106] Calculation process:

[0107] Suppose there are 5 consecutive pixels in the image, and their grayscale values are:

[0108] G1=120, G2=125, G3=130, G4=135, G5=140.

[0109] The standard of pixel spacing being 1 pixel is adopted, that is, p=1.

[0110] Calculate the grayscale difference between each pair of adjacent pixels:

[0111] G2-G1=125-120=5, G3-G2=130-125=5, G4-G3=135-

[0112] 130=5,G5-G4=140-135=5;

[0113] Compute the absolute values of these differences:

[0114] |G2-G1|=5, |G3-G2|=5, |G4-G3|=5, |G5-G4|=5;

[0115] Find the sum of the grayscale value differences of each pair of pixels:

[0116]

[0117] Compute the sum of squares of the margins for each pair of pixels:

[0118]

[0119] Finally, calculate the repeat pitch length:

[0120]

[0121] Interpretation of the results

[0122] The calculated result L rep = 10 represents the length of the repeating pitch within the image region, indicating the periodicity of the grayscale value changes in that region. A value of 10 means that within the selected region, the grayscale change period between pixels is approximately 10 pixels. This indicates that the texture direction of the image has stable repeatability and that these repeating pitches are the main characteristics of this region.

[0123] S303: Based on the texture direction and pitch classification results, statistical processing is performed, and iterative grouping is performed according to the texture main axis angle difference and pitch change trend to generate a closed structure texture direction group;

[0124] First, for each group, the difference in the principal axis angle of the texture direction is calculated. The principal axis angle is a representative value of the texture direction, which is usually determined by calculating the mean of the pixel gradient in that direction. For example, if the texture direction distribution angle range of a certain area is 45 degrees to 55 degrees, the principal axis angle is 50 degrees, and the difference between it and other areas is calculated. If the difference is small (for example, less than 5 degrees), they are classified into the same group. Next, the pitch change trend in each group is statistically processed, that is, the change in pitch length between different areas. By analyzing the pitch change trend, it is determined whether there is a significant pitch change. For example, the trend of the pitch changing from 10 pixels to 20 pixels can be regarded as an obvious change, and then the areas with similar trends are divided into the same group. Through this iterative grouping method, a closed structure texture direction group is finally generated, which contains areas with similar texture directions and pitches. This set of results is helpful for subsequent texture feature analysis.

[0125] See also Figure 5 , the specific steps of S4 are:

[0126] S401: calling the structure contour marked area in the closed structure texture direction group, obtaining the texture main axis direction angle and pitch length in the differential image frame corresponding to the structure area with the same number, calculating the texture main axis direction difference of the area, analyzing the pitch change amplitude, and obtaining the direction angle and pitch difference value;

[0127] First, the structure outline marker regions in the closed structure texture direction group are called upon to extract and obtain the texture principal axis angle and pitch length in the differential image frames corresponding to the structure regions with the same number. This operation first requires identifying and extracting the structure outline marker regions from the image frames. These regions mark the specific boundaries of the structure and ensure that this boundary information is consistent across all frames. Next, the differential image frames are analyzed to extract the texture principal axis angle and pitch length for the structure regions with the same number in these frames. The texture principal axis angle is calculated by analyzing the statistical characteristics of the pixel gradient direction within each structure region, typically calculated based on the grayscale value change of the local pixel. Suppose that within a region, the texture direction of a structure gradually deflects from 0 degrees (horizontally) to 30 degrees from the pixel on the left side of the image, then the principal axis angle is 30 degrees. Next, the pitch length is calculated, which is the distance between the repetitions of texture elements within the region. For example, within this region, the texture repetition period is 15 pixels, indicating a pitch length of 15 pixels. Next, the texture principal axis direction difference of the region is calculated based on the texture principal axis angle and pitch length extracted from the image frames. This difference is calculated from the difference in the main axis angles between adjacent image frames. For example, if the texture angle in one image frame is 30 degrees and in another is 35 degrees, the angle difference is 5 degrees. The pitch length difference is also calculated. If the pitch length in one image frame is 15 pixels and the pitch length in the adjacent frame is 18 pixels, the pitch difference is 3 pixels. Through this process, the direction angle difference and pitch difference values for each structural area can be determined, providing basic data for subsequent image analysis.

[0128] S402: Calculating the offset ratio index of the structural region in adjacent image frames based on the direction angle and the pitch difference value, comparing it with a preset pitch stability reference value, and selecting the region numbers with offset ratios higher than the reference value to obtain the region numbers with abnormal offset ratios;

[0129] The calculation formula of the offset ratio index of the structure area in adjacent image frames is as follows:

[0130]

[0131] Among them, P offset Represents the offset ratio index of the structure area in adjacent image frames, ΔX i represents the horizontal displacement change of the i-th structural region, ΔT irepresents the time interval of the i-th structural region, m is the total number of structural regions, |·| represents the absolute value, and γ is a constant;

[0132] Offset ΔX i :

[0133] The offset represents the horizontal displacement change of the structural area in each frame. It can be calculated by comparing the coordinate position difference of a certain structural area in adjacent frames of the image. In this step, the change in the horizontal position (x coordinate) of each pixel point in the structural area is the offset. The offset can be calculated by the image difference between adjacent frames. For the structural area in consecutive frames, assuming that the horizontal coordinate of a certain area in the first frame is 100 pixels, and the horizontal coordinate of the area in the second frame is 105 pixels, then ΔX1 = |105-100| = 5 pixels.

[0134] Time interval ΔT i :

[0135] The time interval represents the time it takes for the image acquisition system to capture consecutive frames, usually in seconds. The time interval is usually determined by the camera's frame rate or shooting frequency. Assuming the image acquisition frequency is 30 frames / second, the time interval between each frame is ΔT i =1 / 30=0.0333 seconds.

[0136] Constant γ:

[0137] γ is a constant, usually used to avoid the situation where the divisor is zero. In practical applications, the value of γ is generally taken as a very small positive number, usually set to 10 -6 , to ensure calculation stability and avoid division by zero errors in mathematical operations.

[0138] Calculation example:

[0139] Assume that there are 5 structural regions in the image sequence, and their horizontal displacement changes are:

[0140] ΔX1 = 5 pixels, ΔX2 = 6 pixels, ΔX3 = 4 pixels, ΔX4 = 7 pixels, ΔX5 = 5 pixels. The time interval for each frame image is ΔT = 0.0333 seconds.

[0141] Calculate the sum of the absolute values of the offsets:

[0142]

[0143] Compute the sum of time intervals:

[0144]

[0145] Substitute into the formula to calculate the offset ratio P offset :

[0146]

[0147] Result interpretation:

[0148] Calculation result P offset =162.1 represents the offset ratio of the structural region within the image frame. A larger value indicates a greater displacement of the structural region between adjacent frames. This offset ratio can be used to further analyze the stability and anomalies of the structural region. If the offset ratio exceeds a preset threshold, the region can be marked as an anomaly.

[0149] S403: extracting the offset amplitude and distribution trend of the corresponding area according to the offset ratio abnormal area number, and generating a structural texture offset rate list;

[0150] First, based on the number of the area with abnormal offset ratio, the offset amplitude of the corresponding area is extracted from the image. This amplitude refers to the displacement of the area relative to the original position in different frame images. Assuming that a certain area has a displacement of 20 pixels between different frames, the offset amplitude is 20 pixels. Next, analyze the offset distribution trend of these areas to check their offset at different time points or different positions. If the offset trend of a certain area is relatively stable, such as the offset amplitude always remains at around 20 pixels, it means that the offset trend of this area is relatively stable. If the offset amplitude of a certain area fluctuates greatly, such as from 10 pixels to 50 pixels, it is considered that the offset trend of this area is unstable. Based on these data, a list of structural texture offset rates is generated, which records the offset amplitude and distribution trend of each area with abnormal offset ratio, providing basic data for subsequent further analysis.

[0151] See also Figure 6 , the specific steps of S5 are:

[0152] S501: calling the structure number area marked with abnormal offset ratio in the structure texture offset rate list, extracting the boundary point distribution of the corresponding area in the image, calculating the angle change trend between the boundary points, and obtaining the angle change trend value;

[0153] First, the structure-texture offset ratio list, which contains the structure-numbered regions marked with anomalies in the offset ratio, is used to extract the boundary point distribution of the corresponding region in the image. Boundary point distribution refers to the edge of the structured region in the image. Boundary points are pixels on the boundary between the region and its background, typically extracted using edge detection methods such as the Sobel operator or the Canny edge detection algorithm. During execution, the region number with the anomaly in the offset ratio is first obtained from the list. The coordinates of the boundary points within the corresponding region are then extracted. These coordinates reflect the edge morphology of the region. Next, the angle trend between the boundary points is calculated. This angle trend describes the degree of change in boundary shape, specifically the change in directional angle from one boundary point to the next. For example, if the directional angles between two adjacent boundary points are 30 and 50 degrees, respectively, the angle difference between the two points is 20 degrees. Calculating the angle change between multiple points generates an angle trend curve, which reflects the degree of boundary tortuosity. By calculating these angle trends, a specific angle trend value is obtained, which represents the boundary morphology change within the structured region and provides a basis for subsequent analysis.

[0154] S502: Calculating the closed contour shrinkage rate of the continuous region of the boundary points based on the angle change trend value, and comparing the calculated value with a preset standard threshold to screen continuous pixel segments with a direction mutation angle greater than the standard threshold, thereby obtaining regions with a direction mutation angle greater than the threshold;

[0155] Calculate the closed contour shrinkage rate for continuous regions of boundary points. The closed contour shrinkage rate refers to the degree to which the contour of a structural region shrinks as the boundary points move in an image. For example, if a boundary point changes across multiple image frames, causing the boundary of the structural region to shrink, the shrinkage rate can be calculated by calculating the ratio of the initial boundary area to the changed boundary area. Suppose that in a certain image frame, the boundary area of the region is initially 1000 pixels. After several frames of changes, the boundary area is reduced to 800 pixels. The closed contour shrinkage rate of the structure is (1000 - 800) / 1000 = 0.2, which is a 20% shrinkage. The closed contour shrinkage rate of each structural region is calculated and then compared with a preset threshold, assuming the threshold is 15%. When the shrinkage rate exceeds this threshold, the region is considered to have undergone significant structural change. This process can screen out continuous pixel segments with a directional abrupt change angle greater than the threshold. These segments represent regions of significant image change. This allows us to further identify regions with significant boundary abrupt changes, providing key information for subsequent monitoring.

[0156] S503: Mark the location of the area where the sudden change angle is greater than the threshold, perform verification, and generate the vehicle bottom monitoring result;

[0157] First, it is necessary to mark the areas in the image where the directional mutation angle is greater than the preset threshold. For example, if the directional mutation angle of a certain area is 30 degrees and the threshold is set to 20 degrees, then the area meets the conditions and needs to be marked. The position marking process is to mark the position of the mutation area through the image coordinate system. The marked area will clearly indicate the location where the structure has changed. Next, the verification step is carried out. This step mainly uses multiple frames of images to further analyze the marked area to confirm whether the changes in the area are consistent and whether they persist in multiple consecutive image frames to avoid interference from accidental factors. During the verification process, the change trend of the area in different image frames can be calculated to determine whether the mutation exists stably. Finally, after verification, the vehicle bottom monitoring results are generated, and the qualified areas are identified as part of the monitoring results, indicating that the structure of this part of the area has changed significantly, providing a basis for monitoring.

[0158] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A vehicle bottom monitoring method for battery swap stations based on AI vision, characterized in that: The following steps are involved: S1: Obtain image frames of the battery swapping operation area, extract the grayscale distribution of the vehicle bottom contour, calculate the illumination difference and photosensitivity mean change between adjacent frames, identify frames with illumination fluctuation amplitude exceeding the mean as abnormal frames, and generate a photosensitivity distribution enhancement layer sequence; S2: Based on the photosensitive distribution enhanced layer sequence, extract the grayscale jump band of the battery compartment closed groove and the chute hook angle area, calculate the slope extreme value, filter the pixel mean of the non-edge area, remove the noise in the edge error band, and generate a boundary response cleaned image set; S3: calling the boundary response to clean the closed contour pixels in the image set, calculating the boundary closure and pixel continuity, screening the areas with high closure, and generating a closed structure texture direction group; S4: calling the structure area in the closed structure texture direction group, obtaining the texture main axis direction and pitch length, calculating the direction difference and pitch offset between adjacent frames, screening the area numbers with excessive offset, and generating a structure texture offset rate list; S5: Call the abnormal area number in the structure texture offset rate list, extract the boundary point position, calculate the angle change and contour shrinkage, mark the pixels with the direction mutation angle greater than the threshold, and generate the vehicle bottom monitoring result.

2. The method for monitoring the bottom of a vehicle at a battery swap station based on AI vision according to claim 1 is characterized in that: The photosensitive distribution enhancement layer sequence includes a photosensitive abnormality frame index, a grayscale jump distribution layer, and a regional illumination change feature label; the boundary response cleaning image set includes an edge response layer, a non-edge area mean map, and a boundary pixel difference filtering result; the closed structure texture direction group includes a high-closure structure contour, a main axis texture direction classification, and a repeated texture pitch feature; the structure texture offset rate list includes a texture direction change amplitude, a pitch offset ratio parameter, and an abnormal structure number index; the vehicle bottom monitoring result includes a boundary angle change distribution, a contour shrinkage rate index, and a direction mutation area mark.

3. The method for monitoring the bottom of a vehicle at a battery swap station based on AI vision according to claim 1 is characterized in that: The specific steps of S1 are: S101: Obtain a sequence of image frames captured by an image acquisition device in the slide-in section of the battery swapping operation area, extract the grayscale channel pixel distribution of the vehicle bottom contour area in each frame of the image, calculate the light intensity difference based on the grayscale mean and distribution difference in the same area of adjacent image frames, and analyze the sensitivity change rate to obtain the grayscale difference and sensitivity change rate range; S102: Based on the grayscale difference and the sensitivity change rate interval, the illumination change value of each frame image is numerically compared with the average illumination fluctuation amplitude to determine whether the illumination fluctuation amplitude exceeds the average illumination fluctuation threshold, the corresponding frame image is screened as an abnormal frame, and an index set of sensitivity abnormal frames is obtained; S103: Calling the grayscale channel pixel values of the image frames in the photosensitivity abnormality frame index set, detecting the grayscale mutation area in the image, extracting the grayscale jump area by classifying the jump amplitude and distribution morphology, and generating a photosensitivity distribution enhancement layer sequence according to the category mark.

4. The method for monitoring the bottom of a vehicle at a battery swap station based on AI vision according to claim 3 is characterized in that: The specific steps of S2 are: S201: Based on the image frames in the photosensitive distribution enhancement layer sequence, extract the grayscale channel pixel values of the corner area where the edge segment of the battery compartment closed groove and the chute hook are connected, identify the grayscale jump band, and generate a grayscale jump band distribution layer by calculating the grayscale value changes within the image area; S202: Calculate the slope extreme points of the grayscale value sequence based on the grayscale jump band distribution layer, determine the edge response concentration segment in the region, filter the pixel distribution mean of the non-edge region, and eliminate the density difference with the pixel value of the edge response band to obtain the edge response cleaned layer; S203: performing a difference elimination operation according to the edge response cleaning layer, processing the density difference between the boundary pixels and the reference value, screening out the boundary area, and generating a boundary response cleaning image set.

5. The method for monitoring the bottom of a vehicle at a battery swap station based on AI vision according to claim 4 is characterized in that: The slope extreme point calculation formula of the gray value sequence is specifically: Among them, S max Represents the slope extreme point of the gray value sequence, P i Represents the grayscale value of the i-th pixel in the image, P i-1 Represents the grayscale value of the i-1th pixel, Δx i represents the pixel interval between the i-th pixel and the previous pixel, n is the total number of pixels in the grayscale value sequence, and |·| represents the absolute value.

6. The method for monitoring the bottom of a vehicle at a battery swap station based on AI vision according to claim 4 is characterized in that: The specific steps of S3 are: S301: calling the closed contour pixel set in the boundary response cleansing image set, calculating the boundary closure and pixel continuity of each closed structure in the image, screening areas with closure higher than a set ratio, and generating a structure contour unit set; S302: Classifying the texture direction angles of the pixel groups within the structure according to the structure contour unit set, calculating the length of the repeated pitch within each structure region, grouping the texture direction groups according to the pitch variation trend, and obtaining texture direction and pitch classification results; S303: Based on the texture direction and pitch classification results, statistical processing is performed, and iterative grouping is performed according to the texture main axis angle difference and the pitch change trend to generate a closed structure texture direction group.

7. The method for monitoring the bottom of a vehicle at a battery swap station based on AI vision according to claim 6 is characterized in that: The calculation formula for the repeated pitch length in each structural area is specifically: Among them, L rep Represents the length of the repeated pitch in each structural area, G j Represents the grayscale value of the jth pixel in the image, G j-p represents the grayscale value of the jpth pixel in the image, p is a fixed step size, which indicates the pixel interval considered in the grayscale change process, m is the total number of pixels in the image, and |·| indicates taking the absolute value.

8. The method for monitoring the bottom of a vehicle at a battery swap station based on AI vision according to claim 6 is characterized in that: The specific steps of S4 are: S401: calling the structure contour marked area in the closed structure texture direction group, obtaining the texture main axis direction angle and pitch length in the differential image frame corresponding to the structure area with the same number, calculating the texture main axis direction difference of the area, analyzing the pitch change amplitude, and obtaining the direction angle and pitch difference value; S402: Calculating an offset ratio index of a structural region in adjacent image frames based on the direction angle and the pitch difference value, comparing the index with a preset pitch stability reference value, and selecting the region numbers whose offset ratios are higher than the reference value to obtain the region numbers with abnormal offset ratios; S403: According to the number of the region with abnormal offset ratio, the offset amplitude and distribution trend of the corresponding region are extracted to generate a list of structural texture offset rates.

9. The method for monitoring the bottom of a vehicle at a battery swap station based on AI vision according to claim 8 is characterized in that: The calculation formula of the offset ratio index of the structure area in adjacent image frames is specifically: Among them, P offset Represents the offset ratio index of the structure area in adjacent image frames, ΔX i represents the horizontal displacement change of the i-th structural region, ΔT i represents the time interval of the i-th structural region, m is the total number of structural regions, |·| represents the absolute value, and γ is a constant.

10. The method for monitoring the bottom of a vehicle at a battery swap station based on AI vision according to claim 8, characterized in that: The specific steps of S5 are: S501: calling the structure number area marked with abnormal offset ratio in the structure texture offset rate list, extracting the boundary point distribution of the corresponding area in the image, calculating the angle change trend between the boundary points, and obtaining the angle change trend value; S502: Calculating the closed contour shrinkage rate of the continuous region of boundary points based on the angle change trend value, and screening the continuous pixel segments whose direction mutation angle is greater than the standard threshold by comparing the calculated value with a preset standard threshold, to obtain the region where the direction mutation angle is greater than the threshold; S503: Mark the position of the area where the sudden change angle of the direction is greater than the threshold, perform verification, and generate a vehicle bottom monitoring result.

Images