Automobile leather defect detection method and system based on visual detection

Abstract

The invention discloses an automobile leather defect detection method and system based on visual inspection, and relates to the technical field of industrial visual inspection, and the method comprises the steps: reconstructing the three-dimensional shape of a leather surface and generating a three-dimensional point cloud picture by analyzing the parallax relation and illumination direction reflection characteristics among multi-view automobile leather images; automobile leather surface curvature change characteristics of the three-dimensional point cloud picture are extracted, multi-scale texture analysis is carried out, and potential defect areas are identified and marked; through a three-dimensional shape measurement method, defect three-dimensional shape characteristics of the potential defect area are extracted, and defect three-dimensional geometric parameters are calculated; by analyzing defect geometrical characteristics and spatial distribution rules of the defect three-dimensional geometrical parameters and utilizing a preset grading judgment rule to divide defect grades, an automobile leather quality evaluation report containing defect three-dimensional coordinates is generated; according to the method, through combination of curvature-texture multi-scale fusion detection, the recognition capability of complex surface defects is remarkably enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial visual inspection, and in particular to a method and system for detecting automobile leather defects based on visual inspection. Background Art

[0002] In recent years, automotive leather surface defect detection technology has evolved from traditional manual visual inspection to automated inspection methods based on machine vision. Current mainstream technologies primarily utilize 2D image processing algorithms (such as edge detection and texture analysis) and single-view 3D scanning techniques (such as structured light projection and laser triangulation) for defect identification. Multispectral imaging, by analyzing the optical properties of different wavelengths, effectively identifies surface color differences and texture anomalies. Structured light-based 3D reconstruction methods, on the other hand, enable submillimeter surface topography measurement.

[0003] Current automotive leather defect detection technology faces significant limitations in quantifying three-dimensional features. Existing methods primarily rely on two-dimensional image analysis, which can only capture planar projection features of defects (such as area and perimeter), but cannot accurately characterize key three-dimensional geometric parameters such as dent depth and edge abruptness. This 2D quantification approach can lead to the loss of important topographic information, particularly when processing complex, curved leather surfaces. For example, it can't distinguish between a minor surface scratch and a deep crack. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method for detecting automobile leather defects based on visual inspection to solve the problem that the existing technology cannot quantify the three-dimensional geometric features of defects, resulting in insufficient detection accuracy.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In the first aspect, the present invention provides a method for detecting automobile leather defects based on visual inspection, which includes: obtaining multi-perspective automobile leather images through multi-perspective synchronous shooting, and performing dynamic surface focusing to generate clearly focused multi-perspective automobile leather images; reconstructing the three-dimensional morphology of the leather surface by analyzing the parallax relationship and light direction reflection characteristics between the multi-perspective automobile leather images, and generating a three-dimensional point cloud map; extracting the curvature change characteristics of the automobile leather surface in the three-dimensional point cloud map, and performing multi-scale texture analysis to identify and mark potential defect areas; extracting the three-dimensional morphological characteristics of the defects in the potential defect areas through a three-dimensional morphology measurement method, and calculating the three-dimensional geometric parameters of the defects; analyzing the defect geometric characteristics and spatial distribution laws of the three-dimensional geometric parameters of the defects, and dividing the defect levels using preset grading judgment rules, to generate an automobile leather quality assessment report containing the three-dimensional coordinates of the defects.

[0007] As a preferred solution of the automobile leather defect detection method based on visual inspection described in the present invention, the multi-view automobile leather image includes automobile leather images at an orthographic perspective, an oblique perspective, and a grazing perspective.

[0008] As a preferred solution of the automotive leather defect detection method based on visual inspection of the present invention, wherein: the dynamic surface focusing is performed to generate a clear and focused multi-view automotive leather image, the specific steps are as follows: Based on multi-view automotive leather images, the Gaussian curvature and average curvature distribution of the automotive leather surface are calculated, and the focus of the binocular camera is controlled in real time to obtain the locally optimally focused automotive leather image set at each viewpoint. The locally optimally focused automotive leather image sets from each perspective are fused in the frequency domain, and sub-pixel geometric correction is performed through multi-perspective feature matching to generate a multi-perspective automotive leather image set with full field of view clarity. For a multi-view car leather image set with clear full field of view, polarization reflection component separation and light intensity compensation are performed to generate enhanced multi-view car leather images; The enhanced multi-view car leather image is subjected to disparity analysis and the depth distance from the car leather surface to the binocular camera is calculated to generate a clear and focused multi-view car leather image.

[0009] As a preferred embodiment of the automotive leather defect detection method based on visual inspection described in the present invention, the method reconstructs the three-dimensional topography of the leather surface and generates a three-dimensional point cloud image by analyzing the parallax relationship between multi-view automotive leather images and the reflection characteristics of the light direction. The specific steps are as follows: Based on the clearly focused multi-view car leather images, the disparity values ​​between the viewpoints are calculated using the SIFT feature matching algorithm to obtain the disparity matrix with sub-pixel accuracy. Using the disparity matrix combined with the preset illumination direction, the surface normal vector field is reconstructed through photometric stereo vision to generate a three-dimensional gradient map containing the vector inclination and azimuth. The 3D gradient image and the disparity matrix are reconstructed using Poisson method to obtain the initial 3D point cloud image, and outlier filtering and hole repair are performed to generate a 3D point cloud image.

[0010] As a preferred embodiment of the automotive leather defect detection method based on visual inspection of the present invention, the following steps are used to extract the curvature variation characteristics of the automotive leather surface from the three-dimensional point cloud image and perform multi-scale texture analysis to identify and mark potential defect areas: The 3D point cloud image, Gaussian curvature and mean curvature of the car leather surface are combined, and differential geometry parameter mapping and curvature threshold segmentation are performed to generate a curvature distribution map; The curvature distribution map and multi-view car leather images are decomposed by wavelet multi-scale to obtain the texture energy distribution characteristics of each frequency band; The curvature distribution map and texture energy distribution features are weightedly fused at the pixel level to generate a defect probability heat map. Adaptive threshold segmentation is then used to output a three-dimensional coordinate defect bounding box to mark the potential defect area.

[0011] As a preferred embodiment of the automotive leather defect detection method based on visual inspection of the present invention, the three-dimensional morphological features of the potential defect area are extracted by a three-dimensional morphology measurement method, and the three-dimensional geometric parameters of the defect are calculated. The specific steps are as follows: Based on the marked potential defect area, the structured light 3D topography measurement method is used to perform multi-view fringe projection and phase solution to obtain 3D point cloud data of the defect area; Through the surface reconstruction algorithm, the three-dimensional morphological features of the defect are extracted from the three-dimensional point cloud data of the defect area; The three-dimensional point cloud data of the defect area and the three-dimensional morphological features of the defect are fused and calculated in the spatial domain to obtain the three-dimensional geometric parameters of the defect.

[0012] As a preferred embodiment of the method for detecting automobile leather defects based on visual inspection according to the present invention, wherein: The method analyzes the defect geometric characteristics and spatial distribution of the three-dimensional geometric parameters of the defect, and uses preset classification rules to divide the defect levels to generate an automotive leather quality assessment report containing the three-dimensional coordinates of the defect. The specific steps are as follows: Perform spatial statistical analysis based on the three-dimensional geometric parameters of the defect and the three-dimensional point cloud data of the defect area to quantify the defect concentration index; The defect concentration index is scored according to preset grading rules, and combined with the three-dimensional point cloud data of the defect area to generate an automotive leather quality assessment report containing the three-dimensional coordinates of the defects.

[0013] In the second aspect, the present invention provides an automobile leather defect detection system based on visual inspection, including an imaging module, a reconstruction module, an identification module, an analysis module and a generation module; the imaging module is used to obtain multi-perspective automobile leather images through multi-perspective synchronous shooting, and perform dynamic surface focusing to generate clearly focused multi-perspective automobile leather images; the reconstruction module is used to reconstruct the three-dimensional morphology of the leather surface by analyzing the parallax relationship and light direction reflection characteristics between the multi-perspective automobile leather images, and generate a three-dimensional point cloud map; the identification module is used to extract the curvature change characteristics of the automobile leather surface in the three-dimensional point cloud map, and perform multi-scale texture analysis to identify and mark potential defect areas; the analysis module is used to extract the three-dimensional morphological characteristics of the defects in the potential defect area through a three-dimensional morphology measurement method, and calculate the three-dimensional geometric parameters of the defects; the generation module is used to analyze the defect geometric characteristics and spatial distribution laws of the three-dimensional geometric parameters of the defects, and divide the defect levels using preset grading judgment rules to generate an automobile leather quality assessment report containing the three-dimensional coordinates of the defects.

[0014] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the automotive leather defect detection method based on visual inspection as described in the first aspect of the present invention is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the automotive leather defect detection method based on visual inspection as described in the first aspect of the present invention is implemented.

[0016] The beneficial effects of the present invention are as follows: through dynamic surface focusing, a clearly focused multi-perspective image of automobile leather is generated, and clear imaging of the entire field of view of the automobile leather surface is achieved. The optical focusing based on real-time curvature calculation solves the imaging defects of fixed focal length on curved surface materials, and provides a high-quality image data foundation for subsequent three-dimensional reconstruction; combined with curvature-texture multi-scale fusion detection, the geometric features of the three-dimensional point cloud and the texture features of the two-dimensional image are deeply integrated, breaking through the technical limitations of single-modality detection, significantly enhancing the recognition ability of complex surface defects, and achieving improvements in the accuracy and efficiency of automobile leather defect detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 The figure is a flow chart of the automotive leather defect detection method based on visual inspection.

[0019] Figure 2 Schematic diagram of the automotive leather defect detection system based on vision inspection.

[0020] Figure 3 Flowchart for generating 3D point cloud images.

[0021] Figure 4 Flowchart for generating a quality assessment report for automotive leather. DETAILED DESCRIPTION

[0022] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0023] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0025] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a method for detecting automobile leather defects based on visual inspection, comprising the following steps: S1. Acquire multi-view automotive leather images through multi-view synchronous shooting.

[0026] S1.1. The multi-view car leather images include car leather images at an orthographic view, an oblique view, and a grazing view.

[0027] Specifically, the binocular cameras are fixedly installed above the inspection station with an orthogonal viewing angle of 90 degrees, an oblique viewing angle of 45 degrees, and a grazing viewing angle of 15 degrees, respectively. The spacing between the binocular cameras is set to 300 mm for example. The optical axis of the orthogonal binocular camera lens is perpendicular to the surface of the car leather, the angle between the optical axis of the oblique binocular camera and the normal is 45 degrees, and the angle between the optical axis of the grazing binocular camera and the normal is 15 degrees. The three binocular cameras are triggered synchronously by hardware to capture images, and the exposure time is set to 10 ms for example. During the acquisition process, the car leather sample passes through the inspection area at a constant speed of 50 mm / s for an example, and the binocular cameras shoot synchronously at an exemplary frame rate of 30 fps. The binocular cameras simultaneously acquire orthogonal, oblique, and grazing images to form multi-perspective images of the car leather.

[0028] S2. Perform dynamic curved surface focusing to generate a clearly focused multi-view car leather image.

[0029] S2.1. Based on multi-view automotive leather images, calculate the Gaussian curvature and average curvature distribution of the automotive leather surface, and control the focus of the binocular camera in real time to obtain the locally optimally focused automotive leather image set at each viewpoint.

[0030] It should be noted that the expressions for calculating the Gaussian curvature and average curvature distribution of the automobile leather surface based on multi-view automobile leather images are: ; in, is the Gaussian curvature distribution of the car leather surface, is the stretching strength of the car leather surface in the first direction, is the bending strength of the car leather surface in the second direction, is the degree of distortion of the car leather surface in the first and second directions, is the bending strength of the car leather surface in the first direction, is the stretching strength of the car leather surface in the second direction, is the degree of coupling of the car leather surface in the first and second directions; ; in, is the mean curvature distribution, is the grayscale change rate of the first direction of the multi-view car leather image, is the mixed grayscale gradient of the first and second directions of the multi-view car leather image, is the grayscale change rate of the second direction of the multi-view car leather image; Specifically, the grayscale gradient of the input multi-view automotive leather image is calculated pixel by pixel. The Sobel operator is used to obtain the grayscale change rate in the first and second directions of the multi-view automotive leather image. The grayscale gradient distribution characteristics are statistically analyzed within a fixed-size neighborhood window to calculate the elastic strength of the automotive leather surface in the first and second directions, as well as the degree of coupling between the two directions. The local surface equation is fitted using the least squares method to solve the bending strength of the automotive leather surface in the first and second directions, as well as the degree of distortion between the two directions, to obtain the surface feature parameters obtained from the multi-view automotive leather image. Based on the surface feature parameters acquired from multi-view automotive leather images, a mapping of these parameters is performed to obtain a Gaussian curvature distribution reflecting the surface's concave-convex characteristics and an average curvature distribution representing the overall curvature. Based on the resulting curvature distribution, the binocular camera drives a mechanical transmission device to adjust the lens position. During the focusing process, the image clarity changes in local areas are monitored in real time. When the clarity reaches optimal, the focus is locked, ultimately outputting a locally optimally focused automotive leather image set from each viewpoint. Furthermore, the process of determining whether the clarity has reached the optimal state is as follows: during the focusing process, the clarity changes of the local area of ​​the image are continuously monitored, and the gradient energy value of the multi-view automotive leather image is calculated in real time using the Tenengrad gradient function. When the gradient energy value of the multi-view automotive leather image reaches a peak value, exemplarily 15,000-20,000, and begins to decrease, it is determined that the clarity is in the optimal state, and a control signal is immediately sent to lock the current focal length position, keeping the current binocular camera lens position unchanged. The process is independently executed at each focusing position to ensure that the image of each viewpoint reaches the optimal focus state. It should be noted that the expression for calculating the gradient energy value of multi-view car leather images in real time using the Tenengrad gradient function is: ; in, is the gradient energy value of the multi-view car leather image, Is the traversal window All pixel coordinates within is the column coordinate of the multi-view car leather image pixel, is the row coordinate of the multi-view car leather image pixel, is the grayscale of the multi-view car leather image, is the grayscale matrix of the multi-view car leather image.

[0031] S2.2. Frequency domain fusion is performed on the locally optimally focused automobile leather image set at each perspective, and sub-pixel geometric correction is performed through multi-perspective feature matching to generate a multi-perspective automobile leather image set with full field of view clarity.

[0032] Specifically, a fast Fourier transform is performed on the locally optimally focused automobile leather image at each viewing angle to convert it into the frequency domain, a Butterworth filter is used in the frequency domain space to extract the effective frequency band of the locally optimally focused automobile leather image at each viewing angle, the cutoff frequency of the Butterworth filter is set to an exemplary value of 0.6 times the Nyquist frequency, the frequency domain information of the Butterworth filter is fused by weighted superposition, the clarity weight coefficient is dynamically allocated according to the clarity evaluation result of the Butterworth filter, and an inverse Fourier transform is performed on the fused frequency domain data to reconstruct the spatial domain image; Based on the reconstructed spatial domain image, no fewer than 500 matching feature points are extracted between the multi-view automotive leather images. The SIFT feature detection algorithm is used to calculate a 128-dimensional descriptor for each feature point. The initial matching relationship is obtained through the nearest neighbor matching method. Four pairs of matching points are randomly selected to obtain the initial homography matrix. The initial homography matrix is ​​used to test the reprojection error of all matching points. Matching points with an error less than an exemplary 1.5 pixels are identified as inliers and counted. The exemplary random sampling was repeated 2000 times, and the homography matrix with the largest number of inliers was retained as the optimal matching relationship. Based on the optimal matching relationship, sub-pixel image registration was achieved through bicubic interpolation. Histogram matching processing was performed on the locally optimally focused automotive leather images of each perspective after registration to unify the grayscale distribution of each image, and finally a multi-perspective automotive leather image set with clear full-field view was generated.

[0033] S2.3. Perform polarization reflection component separation and light intensity compensation on the multi-view car leather image set with full field of view clarity to generate an enhanced multi-view car leather image.

[0034] Specifically, a rotating polarizer is used to obtain a sequence of multi-view automotive leather image sets with full field of view and different polarization angles. The diffuse reflection component and the specular reflection component are separated by the Stokes vector method, and a polarization threshold is set. The areas with a polarization degree greater than the polarization threshold are determined to be the specular reflection-dominated areas and removed. The physical separation of the diffuse reflection component and the specular reflection component is achieved; the diffuse reflection component is subjected to median filtering; a reference light intensity template of a defect-free automotive leather area is selected to extract the RGB three-channel reference value; the brightness mean is analyzed within a local window, and when the deviation from the brightness mean exceeds the set polarization threshold, the pixel brightness is adjusted to complete the light intensity compensation of the multi-view automotive leather image set with full field of view and clearness; gamma correction is performed on the multi-view automotive leather image set with full field of view and clearness after intensity compensation; the pixel values ​​of the filtered diffuse reflection component multi-view automotive leather image set are subjected to three-segment mapping conversion, and finally an enhanced multi-view automotive leather image with improved defect signal-to-noise ratio is generated; Furthermore, the process of setting the polarization threshold is as follows: the polarization value distribution of each pixel point is calculated based on the Stokes vector method, the polarization difference characteristics between the normal area and the defective area of ​​the leather surface are analyzed, and the critical value for distinguishing between the specular reflection and diffuse reflection components is determined through statistical histogram, and finally the polarization threshold is obtained.

[0035] S2.4. Perform disparity analysis on the enhanced multi-view car leather image and calculate the depth distance from the car leather surface to the binocular camera to generate a clearly focused multi-view car leather image.

[0036] Specifically, SIFT feature points are extracted from the enhanced multi-view automotive leather images of left and right perspectives, and the corresponding feature point relationships are obtained through the normalized cross-correlation matching method, with the matching window size exemplarily set to 15×15 pixels; a semi-global stereo matching algorithm is used to obtain disparity values, and the disparity search range is exemplarily set to 64 levels, with penalty coefficients exemplarily set to P1=8 and P2=32; the disparity values ​​are converted into depth distances based on the baseline distance and focal length parameters of the binocular camera, and the depth calculation accuracy is exemplarily controlled at the 0.1mm level; the enhanced multi-view automotive leather image fusion is guided based on the depth distance map, and the fusion weight is exemplarily distributed according to a Gaussian distribution (σ=5 pixels); unsharp masking is performed on the fused enhanced multi-view automotive leather image, and the enhancement factor is exemplarily set to 0.5, and finally a clearly focused multi-view automotive leather image is output; It should be noted that the specific process of performing unsharp mask processing on the fused enhanced multi-view automotive leather image is to use a Gaussian filter to smooth the enhanced multi-view automotive leather image. The filter size is exemplarily set to 5×5 pixels. The high-frequency component containing leather surface details is obtained by performing a difference operation between the original enhanced multi-view automotive leather image and the smoothed enhanced multi-view automotive leather image, and then superimposed back to the original enhanced multi-view automotive leather image; the pixel value of the superposition result is truncated to ensure that the channel value of the output clear and focused multi-view automotive leather image remains within the range of 0-255; The expression for calculating the depth distance from the car leather surface to the binocular camera is: ; in, is the depth distance from the car leather surface to the binocular camera, is the binocular camera baseline distance, is the disparity value between the left binocular camera image and the right binocular camera image, is the horizontal pixel coordinate difference between the left and right binocular camera images.

[0037] S3. By analyzing the parallax relationship between multi-view automotive leather images and the reflection characteristics of the light direction, the three-dimensional morphology of the leather surface is reconstructed and a three-dimensional point cloud image is generated.

[0038] S3.1. Based on the clearly focused multi-view car leather image, the disparity values ​​between the viewpoints are calculated using the SIFT feature matching algorithm to obtain a disparity matrix with sub-pixel accuracy.

[0039] It should be noted that the expression for calculating the disparity value between viewing angles using the SIFT feature matching algorithm is: ; in, It's a perspective room The disparity value, is the horizontal coordinate of the left view, is the horizontal coordinate of the right view, is the grayscale of the left view, is the grayscale of the right view; Specifically, based on a clearly focused multi-view car leather image, SIFT feature points are extracted from the multi-view car leather image with clear left and right viewpoints, a Gaussian difference pyramid is constructed and extreme points are located in the scale space. SIFT feature point detection sets an exemplary contrast of 0.04 and an exemplary edge value of 10; a 128-dimensional descriptor of the feature point is obtained and a main direction is assigned; The initial left-right view matching relationship was obtained using the bidirectional nearest neighbor matching method. The exemplary distance ratio was set to 0.8. The RANSAC algorithm was then used for optimization. Four pairs of left-right view matching points were randomly selected each time to calculate the fundamental matrix. The number of inliers with a reprojection error of less than 1.5 pixels was counted. After 2000 iterations, the fundamental matrix with the largest number of inliers was retained and used to select the final left-right view matching point pairs. A 5×5 pixel neighborhood is extracted from the matching point locations in the left and right views, grayscale gradient information is calculated, and sub-pixel accurate coordinates are obtained through quadratic surface fitting. The horizontal coordinate difference is calculated by traversing all matching point pairs in the left and right views, and mapped to a matrix of the same size as the left view. Positions without matching points are filled with nearest neighbor interpolation. Apply a 3×3 pixel median filter to smooth the disparity matrix, detect and correct the disparity jump area, perform mirror filling processing on the boundary area, and output the final optimized sub-pixel precision disparity matrix; It should be noted that the process of constructing a Gaussian difference pyramid is as follows: the input image is Gaussian blurred layer by layer, and five blurred images of different scales are generated at each pyramid level using increasing Gaussian kernel parameters; blurred images of adjacent scales are subtracted pixel by pixel to form a differential image group at each level; a multi-level pyramid structure is established through downsampling, and finally a Gaussian difference pyramid is obtained.

[0040] S3.2. Utilize the disparity matrix combined with the preset illumination direction to reconstruct the surface normal vector field through photometric stereo vision and generate a three-dimensional gradient map containing the vector inclination and azimuth.

[0041] Specifically, the disparity matrix is ​​converted into a depth map, and the depth value is obtained by the baseline distance and focal length parameters of the binocular camera; a preset illumination direction vector is used, with exemplary values ​​of 0.2 for the X component, 0.1 for the Y component, and 0.9 for the Z component; the grayscale values ​​of corresponding pixels are extracted from the car leather images collected under different illumination conditions; the surface normal vector is derived based on the relationship between the pixel brightness change and the illumination direction, and the surface normal vector is calculated using the least squares optimization method; the surface normal vector is decomposed into two components, the inclination angle and the azimuth angle, where the inclination angle represents the angle between the normal vector and the vertical direction, and the azimuth angle represents the projection direction of the normal vector on the horizontal plane; the inclination angle and azimuth angle data are smoothed, exemplarily using bilateral filtering with a 3×3 pixel window; and finally, a dual-channel three-dimensional gradient map consisting of the inclination angle and the azimuth angle is output; Furthermore, the process of converting the disparity matrix into a depth map is based on the geometric principles of binocular stereo vision. The pre-calibrated binocular camera baseline distance (an exemplary value is 100 mm) and focal length parameters (an exemplary value is 1200 pixels) are used to obtain the depth value through the depth calculation formula. The depth value is linearly normalized and mapped to the 0-255 grayscale range to generate the final depth map. It should be noted that the preset process of the lighting direction is as follows: in the lighting experiment, an array of point light sources is set up, and a calibration sphere or a calibration object of known geometric shape is placed at the position of the car leather sample; images of the calibration object under different light sources are collected, and the spatial direction vector of each light source is calculated by locating the high light point through mirror reflection; the relative position parameters of the light source and the binocular camera are recorded, including the elevation and azimuth angles; the elevation and azimuth direction vectors are normalized into unit vectors and stored as lighting parameters. An exemplary main light source direction vector is (0.2, 0.1, 0.9); and the angular error is controlled within an exemplary range of ±1°.

[0042] S3.3. Perform Poisson reconstruction on the 3D gradient image and the disparity matrix to obtain the initial 3D point cloud image, and then perform outlier filtering and hole repair to generate a 3D point cloud image.

[0043] Specifically, the inclination data of the 3D gradient map is verified to ensure that the value is within the range of 0 to 90 degrees, the azimuth data is verified to ensure that the value is within the range of 0 to 360 degrees, and outliers are removed; an octree spatial index structure is created, and the voxel resolution is set to an exemplary 0.1 mm. The Poisson equation is discretized in 3D space and iteratively solved using the conjugate gradient method. The isosurface is extracted from the scalar field to generate a triangular mesh; Statistical outlier removal is performed, and the average distance from each coordinate point to an exemplary 30 neighboring coordinate points is obtained to generate a distance distribution histogram. Holes with more than an exemplary number of closed boundary edges exceeding 20 are identified, and a Delaunay triangulation mesh is generated based on the three-dimensional gradient image. New sampling points are uniformly inserted within the mesh at a spacing of 0.83 times the average point spacing of the original three-dimensional gradient image, so that the total number of sampling points reaches 1.2 times that of the original three-dimensional gradient image. A vertex neighborhood with an exemplary radius of 2 mm is defined and Laplace smoothing is performed with an exemplary three-step displacement weight of 0.2. Finally, a three-dimensional point cloud image of the car leather surface is output. It should be noted that the process of creating an octree spatial index structure is as follows: by traversing the X / Y / Z coordinates of all points in the three-dimensional gradient map, respectively extracting the minimum and maximum values ​​of the three dimensions X / Y / Z, and using the coordinates of the six extreme points of X / Y / Z as boundaries to determine the range of the spatial bounding box, the cubic space is equally divided into eight sub-cubes along the X / Y / Z axis; recursively performing the same segmentation operation on each sub-cube until the preset depth level or voxel resolution requirement is reached; establishing a spatial index number for each final level of voxel units, recording the contained point cloud data pointer; constructing parent-child relationship pointers between levels to form a complete tree-like index structure, and finally obtaining an octree spatial index structure.

[0044] S4. Extract the curvature change characteristics of the automotive leather surface from the 3D point cloud image and perform multi-scale texture analysis to identify and mark potential defect areas.

[0045] S4.1. Combine the three-dimensional point cloud image, the Gaussian curvature and the mean curvature of the automobile leather surface, and perform differential geometry parameter mapping and curvature threshold segmentation to generate a curvature distribution map.

[0046] Specifically, based on the Gaussian curvature and average curvature of the automobile leather surface, the Gaussian curvature and average curvature of the automobile leather surface are converted into visual color codes, where the Gaussian curvature is represented by a red-green spectrum and the average curvature is represented by a blue-yellow spectrum; the curvature threshold range is set to distinguish different feature areas, and Gaussian curvature values ​​in the range of -0.5 to +0.5 mm⁻² are exemplarily corresponded to different color intensities; continuous curvature feature areas are identified by the region growing method, and points with an absolute curvature value exceeding 0.3 mm⁻¹ are exemplarily selected as the starting points of region growth; the visualization results of the Gaussian curvature and average curvature of the automobile leather surface are superimposed and fused to form a composite image that comprehensively reflects the concave and convex characteristics of the leather surface; and finally, a curvature distribution map of the automobile leather surface containing color codes and feature area markers is output; It should be noted that the process of setting the curvature threshold is as follows: analyzing the numerical distribution of Gaussian curvature and mean curvature at all points on the surface of the car leather to generate a hyperbolic statistical histogram; using the Otsu automatic threshold calculation method to perform modal separation on the histogram to determine the critical interval of the curvature distribution; and finally outputting the curvature threshold.

[0047] S4.2. Perform wavelet multi-scale decomposition on the curvature distribution map and the multi-view car leather image to obtain the texture energy distribution characteristics of each frequency band.

[0048] Specifically, the curvature distribution map of the bumpy features of the automotive leather is spatially registered with the multi-view automotive leather images of the corresponding viewpoints to ensure accurate pixel position correspondence; the Daubechies wavelet basis function is selected, and the exemplary decomposition layer number is set to 3; a two-dimensional discrete wavelet transform is performed to decompose the LL low-frequency subband and the LH / HL / HH high-frequency subbands; the texture features of the high-frequency subband are extracted at each scale, and the low-frequency subband is further recursively decomposed; the characteristic response values ​​of each frequency band are recorded and normalized to form a multi-scale feature description, and finally the texture energy distribution characteristics of each frequency band are obtained; It should be noted that the specific process of performing two-dimensional discrete wavelet transform decomposition of subbands is as follows: one-dimensional wavelet filtering operations are performed on the registered multi-view automotive leather image along the row and column directions respectively, and row filtering uses a low-pass and high-pass filter group to process each row of the image, and column filtering performs low-pass and high-pass filtering on the row filtering results again; the results of row filtering and column filtering are cross-combined to generate LL low-frequency subbands (row and column low-pass filtering), LH subbands (row low-pass and column high-pass filtering), HL subbands (row high-pass and column low-pass) and HH subbands (row and column high-pass filtering); the size of each subband is reduced to 1 / 4 of the original multi-view automotive leather image, the LL subband retains the main contour information of the multi-view automotive leather image, the LH subband contains horizontal details, the HL subband contains vertical details, and the HH subband contains diagonal details.

[0049] S4.3. Perform pixel-level weighted fusion of the curvature distribution map and the texture energy distribution features to generate a defect probability heat map, and then segment through adaptive threshold to output the three-dimensional coordinate defect bounding box and mark the potential defect area.

[0050] Specifically, the curvature distribution map of the automotive leather concave-convex feature and the texture feature values ​​of the texture energy distribution feature are normalized to unify the numerical range to the interval of 0-1; the exemplary curvature-texture fusion weight coefficients are set to 0.6 for the curvature feature of the automotive leather concave-convex feature and 0.4 for the texture energy distribution feature, and pixel-level weighted summation is performed to generate a fusion feature map; Gaussian smoothing is performed on the fusion feature map, and the kernel size is exemplarily set to 5×5 pixels; the segmentation threshold is automatically calculated using the Otsu algorithm, and the segmentation threshold is exemplarily obtained to be 0.45; a connected domain analysis is performed on the area exceeding the segmentation threshold, and the minimum defect area is exemplarily set to 25 square millimeters; the two-dimensional defect area is mapped back to the three-dimensional point cloud space, and the three-dimensional convex hull vertex coordinates of the defect area are extracted; an axially aligned three-dimensional bounding box is generated according to the vertex coordinates, and the bounding box expansion coefficient is exemplarily taken as 1.1 times; and finally, the three-dimensional coordinate bounding box information of the potential defect area is output to mark the potential defect area.

[0051] S5. Extract the three-dimensional morphological features of the potential defect area through the three-dimensional morphology measurement method and calculate the three-dimensional geometric parameters of the defect.

[0052] S5.1. Based on the marked potential defect area, use the structured light 3D topography measurement method to perform multi-view fringe projection and phase solution to obtain 3D point cloud data of the defect area.

[0053] Specifically, a structured light projection device is set up within the defect area boundary frame to project an exemplary 12-step phase-shifted sinusoidal fringe pattern, with the fringe period exemplary set to 30 pixels; an industrial camera is synchronously triggered to capture the deformed fringe image, with the industrial camera resolution exemplary being 2048×2048 pixels; a phase solution is performed on the captured deformed fringe image, and a four-step phase shift method is used to calculate the wrapped phase map; the phase is unfolded using a multi-frequency heterodyne method, with the frequencies selected exemplary being 1 / 60, 1 / 50, and 1 / 40; the unfolded absolute phase is converted with a pre-calibrated industrial camera-height mapping relationship to obtain three-dimensional height information; the three-dimensional height information is converted into three-dimensional point cloud coordinates by combining the internal and external parameters of the industrial camera; the defect area data captured from multiple perspectives are subjected to ICP registration, with the registration error exemplary being controlled within 0.05 mm; and finally, the three-dimensional point cloud data of the defect area is output; Furthermore, the calibration process for the pre-calibrated industrial camera-height mapping relationship is as follows: using a standard flat calibration plate, precise displacement is performed in the vertical direction within the measurement space with an exemplary 0.1 mm step size, sinusoidal fringes are projected at each height position, and deformed fringe images are acquired; the phase value corresponding to each height is obtained using a four-step phase shift method, and a continuous mapping function is fitted using cubic spline interpolation, achieving an exemplary 0.01 mm calibration accuracy, ultimately generating the industrial camera height-phase mapping relationship; It should be noted that the process of ICP registration of defect area data collected from multiple perspectives is as follows: find spatial corresponding points between the three-dimensional point clouds of the defect area at different perspectives, and remove abnormal correspondences through the RANSAC algorithm; optimize the rigid body transformation parameters to minimize the distance between corresponding points, and gradually align the point clouds during the iteration process; set an exemplary maximum number of iterations to 50 times and an exemplary distance of 0.1 mm; verify that the registration accuracy meets the exemplary 0.05 mm error requirement; and finally obtain a three-dimensional point cloud that integrates multi-perspective data to complete the ICP registration.

[0054] S5.2. Extract the three-dimensional morphological features of the defect from the three-dimensional point cloud data of the defect area through a surface reconstruction algorithm.

[0055] Specifically, the normal vector of the registered three-dimensional point cloud of the defect area is estimated by the moving least squares method, and the neighborhood radius is exemplarily set to 5 mm; the Poisson surface reconstruction algorithm is used to generate a triangular mesh surface, and the reconstruction depth is exemplarily set to 9 layers; morphological feature parameters are extracted from the reconstructed surface, including the vertical distance from the defect depth to the lowest point of the exemplary measurement reference plane, the number of voxels occupied by the defect volume exemplarily calculated by voxelization, the surface area change rate exemplarily comparing the surface area ratio of the defect area to the surrounding normal area, and the curvature mutation index exemplarily counting the proportion of Gaussian curvature points exceeding 0.25; the feature parameters are normalized by the min-max method to generate a defect morphological descriptor containing an exemplary 12-dimensional feature vector; and finally the three-dimensional morphological features of the defect are output.

[0056] S5.3. Perform spatial domain fusion calculation on the three-dimensional point cloud data of the defect area and the three-dimensional morphological features of the defect to obtain the three-dimensional geometric parameters of the defect.

[0057] It should be explained that the expression of the three-dimensional geometric parameters of the defect is obtained by performing spatial domain fusion calculation: ; in, is the three-dimensional geometric parameter of the defect, is the three-dimensional morphological feature weight of the defect, is the weight of the 3D point cloud data in the defect area, is the three-dimensional morphological feature of the defect, It is the three-dimensional point cloud data of the defect area; Specifically, the three-dimensional morphological features of the defect are normalized so that the three-dimensional morphological feature values ​​of each defect are in the range of 0-1; the three-dimensional point cloud data of the defect area is voxelized, and the voxel resolution is exemplarily set to 0.1 mm; the feature weight parameters are set, and the weight of the three-dimensional morphological features of the defect is exemplarily taken as 0.6, and the weight of the three-dimensional point cloud data of the defect area is exemplarily taken as 0.4; a weighted fusion operation is performed on the three-dimensional point cloud data of the defect area and the three-dimensional morphological features of the defect in the spatial domain, and the defect morphological feature values ​​and point cloud density values ​​of the corresponding positions are combined in each voxel unit; the fused three-dimensional morphological features of the defect are extracted, including the maximum depth of the defect, volume deviation, and curvature integral value; the parameter consistency is verified, and the spatial position error is exemplarily controlled within 0.05 mm; and finally the three-dimensional geometric parameters of the defect including dimensional accuracy and shape accuracy are output.

[0058] S6. By analyzing the defect geometric characteristics and spatial distribution patterns of the defect three-dimensional geometric parameters and using preset grading rules to divide the defect levels, an automotive leather quality assessment report containing the defect three-dimensional coordinates is generated.

[0059] S6.1. Perform spatial statistical analysis based on the three-dimensional geometric parameters of the defect and the three-dimensional point cloud data of the defect area to quantify the defect aggregation index.

[0060] Specifically, the 3D point cloud data of the defect area is divided into exemplary 5mm×5mm×5mm cubic grid cells; the number and intensity values ​​of the 3D geometric parameters of the defects in each cubic grid cell are recorded; the Moran's I index is applied to adjacent grid cells to analyze the spatial autocorrelation; the spatial distribution pattern of the defect parameters is identified; an exemplary clustering threshold of 0.7 is used to distinguish significant clustered areas; the volume percentage and average clustering intensity of the clustered areas are calculated; and finally, the defect clustering index is output. Furthermore, the Moran's I index is applied to adjacent grid cells to analyze spatial autocorrelation. The spatial relationship between adjacent grid cells is described using an inverse distance weighting function, with a neighborhood radius set to 15 mm for example. The defect intensity values ​​of each adjacent grid cell are normalized. The parameter similarity between each adjacent grid cell and its neighboring grid cells is analyzed. The Moran's I index results range from -1 to 1, with positive values ​​reflecting defect clustering and negative values ​​reflecting discrete distribution characteristics. The credibility of the spatial pattern is assessed through a significance test. The analysis results, including the degree of spatial autocorrelation and distribution characteristics, are generated. It should be noted that the process of setting the aggregation threshold is: analyzing the numerical distribution of defect aggregation in historical detection data and drawing the aggregation probability density curve; observing the position of the inflection point of the curve and determining the mutation range of the aggregation value; combining the actual requirements of defect sensitivity in specific application scenarios; and finally obtaining the aggregation threshold through statistical distribution characteristics and engineering experience verification.

[0061] S6.2. Score the defect concentration index according to the preset grading rules, and combine it with the three-dimensional point cloud data of the defect area to generate an automotive leather quality assessment report containing the three-dimensional coordinates of the defects.

[0062] Specifically, the defect concentration index is divided into five exemplary levels according to preset grading rules, with 0-0.3 being level A, 0.3-0.5 being level B, 0.5-0.7 being level C, 0.7-0.9 being level D, and 0.9-1.0 being level E. Spatial coordinate information is extracted from the three-dimensional point cloud data of each defect area, including exemplary X / Y / Z axis coordinate ranges and center of mass positions. The quality score of each defect area is annotated according to the grade classification results. The three-dimensional geometric parameters of the defects, the concentration score, and the coordinate information are integrated. A structured report document is generated, including a defect distribution diagram, a parameter statistics table, and coordinate details. Finally, an automotive leather quality assessment report containing the three-dimensional coordinates of the defects is output. It should be noted that the preset process of the grading rules is: analyze the numerical distribution characteristics of the defect concentration index in the historical inspection data, and count the frequency of occurrence of numerical values ​​in different intervals; observe the impact of different numerical intervals on the product qualification rate in the actual production environment; verify the rationality of each grade interval through batch test data; and finally determine the defect concentration grading rules that include 5 exemplary grades (0-0.3 is grade A, 0.3-0.5 is grade B, 0.5-0.7 is grade C, 0.7-0.9 is grade D, and 0.9-1.0 is grade E).

[0063] This embodiment also provides an automotive leather defect detection system based on visual inspection, comprising: an imaging module, a reconstruction module, a recognition module, an analysis module, and a generation module; An imaging module is used to obtain multi-perspective automotive leather images through multi-perspective synchronous shooting and perform dynamic curved surface focusing to generate clear and focused multi-perspective automotive leather images; The reconstruction module is used to reconstruct the three-dimensional topography of the leather surface by analyzing the parallax relationship between multi-view automotive leather images and the reflection characteristics of the light direction, generating a three-dimensional point cloud image; The recognition module is used to extract the curvature variation characteristics of the automotive leather surface from the 3D point cloud image and perform multi-scale texture analysis to identify and mark potential defect areas; An analysis module is used to extract the three-dimensional morphological features of the potential defect area and calculate the three-dimensional geometric parameters of the defect through a three-dimensional morphology measurement method; The generation module is used to analyze the defect geometric characteristics and spatial distribution laws of the defect three-dimensional geometric parameters, and divide the defect levels using preset grading judgment rules to generate an automotive leather quality assessment report containing the defect three-dimensional coordinates.

[0064] This embodiment also provides a computer device, which is suitable for the case of a method for detecting automobile leather defects based on visual inspection, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the method for detecting automobile leather defects based on visual inspection proposed in the above embodiment.

[0065] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0066] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the automotive leather defect detection method based on visual inspection as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0067] In summary, the present invention achieves full-field clear imaging of the automotive leather surface by: dynamic surface focusing to generate clearly focused multi-perspective images of automotive leather; optical focusing based on real-time curvature calculation solves the imaging defects of fixed focal length on curved surface materials, and provides a high-quality image data foundation for subsequent three-dimensional reconstruction; combined with curvature-texture multi-scale fusion detection, the geometric features of the three-dimensional point cloud and the texture features of the two-dimensional image are deeply integrated, breaking through the technical limitations of single-modality detection, significantly enhancing the recognition ability of complex surface defects, and achieving improvements in the accuracy and efficiency of automotive leather defect detection.

[0068] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for detecting automotive leather defects based on visual inspection, characterized by: include, Acquire multi-perspective car leather images through multi-perspective synchronous shooting, and perform dynamic curved surface focusing to generate clear and focused multi-perspective car leather images; By analyzing the parallax relationship between multi-view automotive leather images and the reflection characteristics of the light direction, the three-dimensional topography of the leather surface is reconstructed and a three-dimensional point cloud image is generated; Extract the curvature variation characteristics of the automotive leather surface from the 3D point cloud image and perform multi-scale texture analysis to identify and mark potential defect areas; Through the three-dimensional morphology measurement method, the three-dimensional morphological features of the potential defect area are extracted and the three-dimensional geometric parameters of the defect are calculated; By analyzing the defect geometric characteristics and spatial distribution patterns of the defect's three-dimensional geometric parameters and using preset grading rules to divide the defect levels, an automotive leather quality assessment report containing the defect's three-dimensional coordinates is generated.

2. The method for detecting automotive leather defects based on visual inspection according to claim 1, wherein: The multi-view car leather image includes car leather images at an orthographic perspective, an oblique perspective, and a grazing perspective.

3. The method for detecting automotive leather defects based on visual inspection according to claim 2, wherein: The specific steps of performing dynamic curved surface focusing to generate a clear and focused multi-view car leather image are as follows: Based on multi-view automotive leather images, the Gaussian curvature and average curvature distribution of the automotive leather surface are calculated, and the focus of the binocular camera is controlled in real time to obtain the locally optimally focused automotive leather image set at each viewpoint. The locally optimally focused automotive leather image sets from each perspective are fused in the frequency domain, and sub-pixel geometric correction is performed through multi-perspective feature matching to generate a multi-perspective automotive leather image set with full field of view clarity. For a multi-view car leather image set with clear full field of view, polarization reflection component separation and light intensity compensation are performed to generate enhanced multi-view car leather images; The enhanced multi-view car leather image is subjected to disparity analysis and the depth distance from the car leather surface to the binocular camera is calculated to generate a clear and focused multi-view car leather image.

4. The method for detecting automotive leather defects based on visual inspection according to claim 3, wherein: The method reconstructs the three-dimensional morphology of the leather surface and generates a three-dimensional point cloud image by analyzing the parallax relationship between multi-view automotive leather images and the reflection characteristics of the light direction. The specific steps are as follows: Based on the clearly focused multi-view car leather images, the disparity values ​​between the viewpoints are calculated using the SIFT feature matching algorithm to obtain the disparity matrix with sub-pixel accuracy. Using the disparity matrix combined with the preset illumination direction, the surface normal vector field is reconstructed through photometric stereo vision to generate a three-dimensional gradient map containing the vector inclination and azimuth. The 3D gradient image and the disparity matrix are reconstructed using Poisson method to obtain the initial 3D point cloud image, and outlier filtering and hole repair are performed to generate a 3D point cloud image.

5. The method for detecting automotive leather defects based on visual inspection according to claim 4, wherein: The specific steps of extracting the curvature change characteristics of the automobile leather surface from the three-dimensional point cloud image and performing multi-scale texture analysis to identify and mark potential defect areas are as follows: The 3D point cloud image, Gaussian curvature and mean curvature of the car leather surface are combined, and differential geometry parameter mapping and curvature threshold segmentation are performed to generate a curvature distribution map; The curvature distribution map and multi-view car leather images are decomposed by wavelet multi-scale to obtain the texture energy distribution characteristics of each frequency band; The curvature distribution map and texture energy distribution features are weightedly fused at the pixel level to generate a defect probability heat map. Adaptive threshold segmentation is then used to output a three-dimensional coordinate defect bounding box to mark the potential defect area.

6. The method for detecting automotive leather defects based on visual inspection according to claim 5, wherein: The three-dimensional morphological characteristics of the potential defect area are extracted by the three-dimensional morphology measurement method, and the three-dimensional geometric parameters of the defect are calculated. The specific steps are as follows: Based on the marked potential defect area, the structured light 3D topography measurement method is used to perform multi-view fringe projection and phase solution to obtain 3D point cloud data of the defect area; Through the surface reconstruction algorithm, the three-dimensional morphological features of the defect are extracted from the three-dimensional point cloud data of the defect area; The three-dimensional point cloud data of the defect area and the three-dimensional morphological features of the defect are fused and calculated in the spatial domain to obtain the three-dimensional geometric parameters of the defect.

7. The method for detecting automotive leather defects based on visual inspection according to claim 6, wherein: The method analyzes the defect geometric characteristics and spatial distribution of the three-dimensional geometric parameters of the defects, and uses preset classification rules to divide the defect levels to generate an automotive leather quality assessment report containing the three-dimensional coordinates of the defects. The specific steps are as follows: Perform spatial statistical analysis based on the three-dimensional geometric parameters of the defect and the three-dimensional point cloud data of the defect area to quantify the defect concentration index; The defect concentration index is scored according to preset grading rules, and combined with the three-dimensional point cloud data of the defect area to generate an automotive leather quality assessment report containing the three-dimensional coordinates of the defects.

8. A system for detecting defects in automotive leather based on visual inspection, based on the method for detecting defects in automotive leather based on visual inspection according to any one of claims 1 to 7, characterized in that: Including, imaging module, reconstruction module, recognition module, analysis module and generation module; An imaging module is used to obtain multi-perspective automotive leather images through multi-perspective synchronous shooting and perform dynamic curved surface focusing to generate clear and focused multi-perspective automotive leather images; The reconstruction module is used to reconstruct the three-dimensional topography of the leather surface by analyzing the parallax relationship between multi-view automotive leather images and the reflection characteristics of the light direction, generating a three-dimensional point cloud image; The recognition module is used to extract the curvature variation characteristics of the automotive leather surface from the 3D point cloud image and perform multi-scale texture analysis to identify and mark potential defect areas; An analysis module is used to extract the three-dimensional morphological features of the potential defect area and calculate the three-dimensional geometric parameters of the defect through a three-dimensional morphology measurement method; The generation module is used to analyze the defect geometric characteristics and spatial distribution laws of the defect three-dimensional geometric parameters, and divide the defect levels using preset grading judgment rules to generate an automotive leather quality assessment report containing the defect three-dimensional coordinates.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the automobile leather defect detection method based on visual inspection according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the automotive leather defect detection method based on visual inspection according to any one of claims 1 to 7 are implemented.

Images