A method for dimensional correction of automobile parts based on image processing

Abstract

The present invention relates to the technical field of image data processing, and more specifically to a method for calibrating the dimensions of automotive parts based on image processing. The method comprises: obtaining the thickness of a rotating plate, an image to be detected, and a standard image of the rotating plate; extracting edge lines in the image to be detected using an improved polynomial interpolation sub-pixel edge positioning algorithm, and performing corrections if there is a deviation from the edge lines of the standard image; wherein the grayscale sampling direction in the algorithm comprises a preset grayscale sampling direction and a deviation direction, the deviation direction being calculated based on the rebound factor of the edge pixel point, the rebound constant of multiple preset direction templates, and directional angle information. The rebound factor is the ratio of the deformation degree of the edge point to the thickness of the rotating plate, and the rebound constant represents the offset angle of the edge point. The present invention solves the problem of low detection accuracy of existing algorithms.

Description

Technical Field

[0001] The present invention relates to the technical field of image data processing, and more particularly to a method for correcting the size of automobile parts based on image processing. Background Art

[0002] In modern automotive manufacturing and quality control, the precise measurement and correction of part dimensions are critical to ensuring assembly accuracy and improving product consistency and reliability. Traditional dimensional inspection methods rely primarily on manual measurement, contact measuring instruments, or single-view image measurement systems. However, these methods have numerous shortcomings in practical applications. For example, contact measurement can cause wear to part surfaces, manual measurement is inefficient and susceptible to subjective interference, and single-view image measurement often fails to provide complete geometric information when faced with complex surfaces or occlusions. This results in low measurement accuracy and an inability to effectively meet the online measurement needs of complex-shaped automotive parts.

[0003] Polynomial interpolation sub-pixel edge positioning algorithms are particularly widely used in the field of automotive parts inspection. For example, when inspecting the edge contours of high-precision structural components such as rotating plates, gears, crankshafts, and camshafts, this algorithm can effectively extract boundary curves and achieve micron-level error control. In complex curved parts such as headlight housings and rearview mirror brackets, irregular edges can be accurately modeled and dimensionally corrected. Furthermore, this algorithm is also applicable to automated quality control processes such as weld spot inspection, sealing strip fit confirmation, and gluing trajectory consistency testing, significantly improving the robustness and practicality of image measurement systems in real-world industrial scenarios.

[0004] To improve edge localization accuracy, academia and industry have extensively studied various sub-pixel edge localization algorithms. Among them, polynomial interpolation has attracted much attention due to its advantages, including a stable mathematical model, high fitting accuracy, and high computational efficiency. The polynomial interpolation sub-pixel edge localization algorithm constructs a grayscale variation curve within the neighborhood of pixel-level edge points and fits it using a polynomial function to determine the precise location of the grayscale extreme points, thereby obtaining edge coordinate information with higher resolution than the original resolution. However, traditional algorithms typically use a fixed gradient direction for grayscale sampling, which cannot flexibly handle situations where the edge direction changes significantly or deformation exists, resulting in low detection accuracy. Summary of the Invention

[0005] In order to solve the problem of low detection accuracy raised in the above background technology, the present invention provides the following solution.

[0006] The present invention provides a method for correcting the size of automobile parts based on image processing, comprising: obtaining the thickness of a rotating plate in an automobile part, an image to be detected, and a standard image of the rotating plate; obtaining an edge line in the image to be detected by using an improved sub-pixel edge positioning algorithm based on polynomial interpolation, and performing correction if there is a deviation between the edge line in the image to be detected and the edge line in the standard image; wherein the improved sub-pixel edge positioning algorithm based on polynomial interpolation includes a grayscale sampling direction, the grayscale sampling direction being the sum of a preset grayscale sampling direction and a deviation direction; the deviation direction for: , For the The normalized rebound factor of edge pixels, For the The first The rebound constant after normalization of edge pixels, For the The angle direction of the preset direction template, is the total number of preset direction templates, For the The preset grayscale sampling direction of the edge pixel point; the rebound factor is The ratio of the deformation degree of the edge pixel point to the thickness of the rotating plate; the rebound constant represents the The offset angle of the edge pixels.

[0007] This technical solution, through an improved sub-pixel edge localization algorithm based on polynomial interpolation, can more accurately locate edge lines in the image being inspected and effectively correct any deviations from the standard image. By adjusting the grayscale sampling direction and introducing a rebound factor, the accuracy of edge localization is further improved, adapting to the deformation and thickness changes of the rotating plate, and enhancing adaptability and robustness to complex edges. The calculation of the rebound constant accurately characterizes edge deviations, providing a more detailed analysis of morphological changes, thereby improving the accuracy and reliability of automotive part dimensional calibration.

[0008] Furthermore, the rebound constant is specifically: The edge pixel is the center, along the The gradient direction of the edge pixel point is obtained by setting a sequence composed of the gradient amplitudes of multiple consecutive pixel points, and the peak point in the sequence is obtained by using the peak detection algorithm. The number of pixels between the edge pixel point and the peak point in the change sequence is taken as the The rebound constant of the edge pixel.

[0009] This technical solution accurately quantifies the rebound constant of edge pixels by acquiring a sequence of gradient amplitudes for multiple consecutive pixels along the gradient direction of the edge pixel and using a peak detection algorithm to identify the peak point. By calculating the number of pixels between the edge pixel and the peak point, it effectively reflects subtle changes in edge position and morphological characteristics, providing a reliable basis for further edge location and image correction.

[0010] Furthermore, the degree of deformation is specifically: obtaining the The edge pixel point is a plurality of preset edge pixel points adjacent to each other on its edge line, and a curve fitting algorithm is used to obtain an adjacent fitting curve of the plurality of preset edge pixel points, and the curvature of the adjacent fitting curve is used as the first The deformation degree of each edge pixel.

[0011] This technical solution accurately characterizes the local geometric features of an edge by acquiring multiple adjacent pixels along the edge line and performing curve fitting. Using the curvature of the fitted curve as the degree of deformation of the edge pixels, it helps identify small but significant local deformations in the image. This not only improves sensitivity to edge morphological changes but also enhances the accuracy and stability of sub-pixel edge localization, providing a reliable deformation basis for subsequent deviation direction calculation and image accuracy correction.

[0012] Furthermore, the image to be detected corresponding to the preset direction template is specifically: a convolution operation is performed on the image to be detected of the rotating plate using multiple convolution kernels of the same size and different directions, wherein each convolution kernel corresponds to a preset direction template, thereby obtaining images to be detected corresponding to multiple preset direction templates.

[0013] The above technical solution can enhance the edge response ability of the image in different directions by performing convolution operations on the image to be detected of the rotating plate using multiple convolution kernels in different directions, so that the edge features in the image can be fully extracted in multiple directions, effectively overcoming the problem of omission or inaccurate recognition when extracting edges in a single direction, thereby improving the comprehensiveness and robustness of edge detection, and providing a richer and more stable directional information foundation for subsequent edge fitting and sub-pixel precise positioning.

[0014] Furthermore, the total number of the preset direction templates is 8, and the corresponding angle directions are 0 degrees, 45 degrees, 90 degrees, 135 degrees, 180 degrees, 225 degrees, 270 degrees and 315 degrees.

[0015] Furthermore, a CCD camera or a CMOS camera is used to obtain an image to be tested and a standard image of the rotating plate in the automobile part, and a laser scanning profilometer is used to obtain the thickness of the rotating plate in the automobile part.

[0016] Furthermore, the image to be detected and the standard image of the rotating plate are subjected to denoising and grayscale processing.

[0017] The above technical solution helps to eliminate noise interference caused by ambient lighting, sensor interference or texture details in the image by denoising and grayscale processing the image to be detected and the standard image of the rotating plate, making the image information purer and more unified. At the same time, it reduces the interference of the color channel on edge detection and contour extraction, improves the stability of image processing and the accuracy of edge positioning, and thus provides more reliable basic data for subsequent size correction.

[0018] Furthermore, the denoising is median filtering or Gaussian filtering.

[0019] Furthermore, an edge detection algorithm is used to obtain edge pixel points in the image to be detected of the rotating plate.

[0020] Furthermore, the edge detection algorithm is a Sobel edge detection algorithm or a Canny edge detection algorithm.

[0021] The beneficial effects of the present invention are:

[0022] By incorporating an improved sub-pixel edge location algorithm based on polynomial interpolation, this invention enables high-precision correction of automotive part dimensions. This is particularly effective in compensating for image edge location deviations when processing complex components such as rotating plates. By incorporating the calculation of the rebound factor and rebound constant, the accuracy and robustness of the correction are further enhanced, taking into account part deformation and thickness differences. Furthermore, through denoising, grayscale processing, edge detection, and the application of preset directional templates, image quality and edge recognition are optimized, making the entire correction process more accurate and reliable, significantly improving the manufacturing precision and quality control of automotive parts. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 FIG. 4 is a flow chart schematically illustrating a method for correcting automobile part size based on image processing according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] An embodiment of a method for calibrating automobile part dimensions based on image processing.

[0025] like Figure 1 As shown in FIG. 1 , a flow chart of a method for correcting automobile part dimensions based on image processing according to an embodiment of the present invention includes the following steps:

[0026] S1: Obtain the thickness of the rotating plate in the automotive part, the image to be tested, and the corresponding standard image.

[0027] In a preferred embodiment, to accurately inspect the rotating plate, a key component in automotive parts, a high-resolution CCD or CMOS camera is first used to capture an image of the rotating plate in its intended state. To ensure comparability and accuracy during the inspection process, a corresponding standard image is also acquired. This standard image represents the rotating plate in a defect-free, structurally intact state that meets factory standards, serving as a reference for subsequent image comparison and analysis.

[0028] In addition to acquiring image information, to further enhance the comprehensiveness and accuracy of inspection, a laser scanning profilometer is used to simultaneously collect thickness information of the rotating plate. This laser scanning profilometer utilizes a non-contact measurement principle, capable of real-time scanning and constructing thickness profiles of the rotating plate along different cross sections. This thickness information can not only be used to detect deformation, wear, dents, or structural unevenness of the rotating plate during manufacturing or long-term use, but can also be combined with image information for multimodal fusion analysis, enabling dual quality assessment of both structure and appearance.

[0029] In another preferred embodiment, to ensure the accuracy and stability of image feature extraction, both the captured image to be detected and the standard image are preprocessed, specifically including image denoising and grayscale processing. The image denoising process mainly uses a median filter or a Gaussian filter algorithm. Among them, the median filter can effectively suppress the salt and pepper noise caused by external interference and maintain the clarity of the image edge. It is particularly suitable for detecting image scenes with high-frequency isolated points. The Gaussian filter, on the other hand, smoothes the entire image by constructing a Gaussian kernel function. It is suitable for reducing local fluctuations in the image caused by illumination changes or sensor noise.

[0030] After denoising, the image is grayscaled. Grayscale conversion is the process of converting a color image into a single-channel grayscale image. This simplifies the complexity of subsequent image processing while highlighting the texture features and edge contours of the rotating plate, thereby improving the stability and accuracy of algorithms such as edge detection and region segmentation.

[0031] S2: Use the improved sub-pixel edge location algorithm based on polynomial interpolation to extract the edge lines in the image to be detected.

[0032] In one embodiment, the improved sub-pixel edge location algorithm based on polynomial interpolation includes a grayscale sampling direction, where the grayscale sampling direction is the sum of a preset grayscale sampling direction and a deviation direction;

[0033] The image to be detected The deviation direction of edge pixels for: , For the The normalized rebound factor of edge pixels, For the The first The rebound constant after normalization of edge pixels, For the The angle direction of the preset direction template, is the total number of preset direction templates, For the The preset grayscale sampling direction of edge pixels;

[0034] By introducing grayscale sampling directions that include deviation directions and comprehensively considering multi-directional image response characteristics, the algorithm can more accurately reflect the true offset trend of edge pixels. Combining the image response of each preset directional template with the corresponding angle information effectively enhances sensitivity to local edge variations and improves the accuracy of sub-pixel edge positioning. By normalizing the rebound factor and rebound constant, the algorithm can also adapt to rotating plates of varying thickness and deformation, enhancing the algorithm's adaptability to complex edge structures and further improving the robustness and accuracy of image resizing.

[0035] The total number of the preset direction templates can be 8, and the corresponding angle directions are 0 degrees, 45 degrees, 90 degrees, 135 degrees, 180 degrees, 225 degrees, 270 degrees and 315 degrees. Of course, they can also be set according to actual conditions.

[0036] An edge detection algorithm is used to obtain edge pixel points in the image to be detected of the rotating plate. The edge detection algorithm is a Sobel edge detection algorithm or a Canny edge detection algorithm.

[0037] By introducing a classic edge detection algorithm to extract edge pixels from the image to be inspected, the contour of the rotating plate can be effectively highlighted, background interference can be reduced, and the accuracy and stability of subsequent edge positioning and deformation analysis operations can be improved. Especially in cases of high image complexity or unclear edge transitions, the clarity and continuity of edge extraction can be improved, providing a reliable image foundation for high-precision dimensional correction.

[0038] The rebound constant is specifically: The edge pixel is the center, along the The gradient direction of the edge pixel point is obtained by setting a sequence composed of the gradient amplitudes of multiple consecutive pixel points, and the peak point in the sequence is obtained by using the peak detection algorithm. The number of pixels between the edge pixel point and the peak point in the change sequence is taken as the The rebound constant of edge pixels.

[0039] By extracting the gradient amplitude of a series of consecutive pixels along the gradient direction of the edge pixel and using a peak detection algorithm to identify the location of significant changes, the distance between the edge pixel and the area of strong change is quantified and used as the pixel's rebound constant. This processing method effectively reflects the relationship between the intensity and distance of changes in the local edge structure of the image, helping to capture edge deviation trends caused by blur, deformation, or interference. Introducing the rebound constant as a factor representing the deviation angle makes the sub-pixel positioning process more sensitive to changes in local image details, thereby improving the accuracy and stability of edge correction.

[0040] The degree of deformation is specifically: obtaining the The edge pixel point is a plurality of preset edge pixel points adjacent to each other on its edge line, and a curve fitting algorithm is used to obtain an adjacent fitting curve of the plurality of preset edge pixel points, and the curvature of the adjacent fitting curve is used as the first The deformation degree of each edge pixel.

[0041] By selecting edge pixels and their adjacent pixels along the edge line and calculating the curvature of the local fitting curve using a curve fitting algorithm, the local variation of the edge is effectively quantified, thereby accurately characterizing the deformation characteristics of the edge pixels. This approach can more sensitively reflect edge fluctuations caused by deformation, manufacturing errors, or shooting angles, making subsequent deviation direction calculations more accurate and reliable. By incorporating local geometric features into the edge localization process, the edge adjustment effect with sub-pixel accuracy is improved.

[0042] The image to be detected corresponding to the preset direction template is specifically: a convolution operation is performed on the image to be detected of the rotating plate using multiple convolution kernels of the same size and different directions, wherein each convolution kernel corresponds to a preset direction template, thereby obtaining images to be detected corresponding to multiple preset direction templates.

[0043] By introducing multiple convolution kernels with different orientations and performing convolution operations on the image to be detected, edge response information can be extracted from the image in different directions, thereby more comprehensively capturing the local feature changes of the rotating plate edge. Compared with traditional edge detection methods that only use fixed orientation, this method improves the directional sensitivity and robustness of edge extraction, making subsequent sub-pixel edge positioning more accurate. Especially in images with complex textures or lighting changes, it can effectively improve positioning accuracy and the reliability of edge line comparison with standard images.

[0044] S3: Compare the edge lines in the image to be detected with those in the standard image, and perform correction if there is any deviation.

[0045] In one embodiment, a target image and a reference image of the rotating plate are first acquired. Edge line information is extracted using an improved sub-pixel edge location algorithm based on polynomial interpolation. Subsequently, edge lines in the target image are compared point by point with corresponding edge lines in the reference image. If any deviation in edge contour position or morphology is found, the edge lines in the target image are adaptively corrected based on the offset direction and magnitude to more closely match the edge contour of the reference image.

[0046] The above solution effectively reduces recognition errors caused by image shooting angle, lighting differences or slight deformation of the workpiece, thereby improving the accuracy and reliability of automotive part dimensional detection and ensuring the precision requirements of subsequent assembly and quality control links.

[0047] The solution of the present invention accurately locates the edge of the rotating plate of automotive parts through an improved sub-pixel edge positioning algorithm based on polynomial interpolation, and corrects any edge deviations, thereby improving detection accuracy. The rebound factor and rebound constant are also used to reflect deformation and offset, further enhancing adaptability to rotating plate deformation. The use of various image processing techniques, such as denoising, grayscale processing, and edge detection, ensures high-quality image input and effectively improves the reliability of image matching. The overall solution enhances the accuracy, robustness, and automation of the detection process, providing strong technical support for high-precision dimensional correction of automotive parts.

[0048] In the description of this specification, "multiple" and "several" mean at least two, such as two, three or more, unless otherwise clearly defined.

[0049] While several embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Numerous modifications, variations, and alternatives will occur to those skilled in the art without departing from the concept and spirit of the present invention. It should be understood that various alternatives to the embodiments of the present invention described herein may be employed in practicing the present invention.

Claims

1. A method for calibrating automobile parts size based on image processing, characterized in that: include: Obtain the thickness of the rotating plate in the automotive parts, the image to be tested and the standard image of the rotating plate; Obtaining edge lines in the image to be detected using an improved sub-pixel edge location algorithm based on polynomial interpolation, and correcting the edge lines in the image to be detected if there is a deviation between the edge lines in the image to be detected and the edge lines in the standard image; The improved sub-pixel edge location algorithm based on polynomial interpolation includes a grayscale sampling direction, which is the sum of a preset grayscale sampling direction and a deviation direction; The deviation direction for: , For the The normalized rebound factor of edge pixels, For the The first The rebound constant after normalization of edge pixels, For the The angle direction of the preset direction template, is the total number of preset direction templates, For the The preset grayscale sampling direction of edge pixels; the rebound constant, specifically: First The edge pixel is the center, along the The gradient direction of the edge pixel point is obtained by setting a sequence composed of the gradient amplitudes of multiple consecutive pixel points, and the peak point in the sequence is obtained by using the peak detection algorithm. The number of pixels between the edge pixel point and the peak point in the sequence is taken as the The rebound constant of edge pixels; The rebound factor is The ratio of the deformation degree of the edge pixel point to the thickness of the rotating plate; the rebound constant represents the The offset angle of the edge pixel points; the degree of deformation, specifically: Get the The edge pixel point is located on the edge line of the preset multiple edge pixel points, and the adjacent fitting curves of the preset multiple edge pixel points are obtained by using the curve fitting algorithm. The curvature of the adjacent fitting curve is used as the first The deformation degree of each edge pixel.

2. The method for calibrating automobile parts size based on image processing according to claim 1, characterized in that: The image to be detected corresponding to the preset direction template is specifically: A convolution operation is performed on the image to be detected of the rotating plate using multiple convolution kernels of the same size and different directions, wherein each convolution kernel corresponds to a preset direction template, thereby obtaining images to be detected corresponding to multiple preset direction templates.

3. The method for calibrating automobile parts size based on image processing according to claim 1, characterized in that: The total number of preset direction templates is 8, and the corresponding angle directions are 0 degrees, 45 degrees, 90 degrees, 135 degrees, 180 degrees, 225 degrees, 270 degrees and 315 degrees.

4. The method for calibrating automobile parts size based on image processing according to claim 1, characterized in that: A CCD camera or a CMOS camera is used to obtain the image to be tested and the standard image of the rotating plate in the automobile part, and a laser scanning profilometer is used to obtain the thickness of the rotating plate in the automobile part.

5. The method for calibrating automobile parts size based on image processing according to claim 1, characterized in that: The image to be detected and the standard image of the rotating plate are subjected to denoising and grayscale processing.

6. The method for calibrating automobile parts size based on image processing according to claim 5, characterized in that: The denoising is performed by median filtering or Gaussian filtering.

7. The method for calibrating automobile parts size based on image processing according to claim 1, characterized in that: The edge detection algorithm is used to obtain edge pixel points in the image to be detected of the rotating plate.

8. The method for calibrating automobile parts size based on image processing according to claim 7, characterized in that: The edge detection algorithm is a Sobel edge detection algorithm or a Canny edge detection algorithm.

Images