CN-121998950-A - Identification method and system for automobile welding spot stud, electronic equipment and storage medium

CN121998950ACN 121998950 ACN121998950 ACN 121998950ACN-121998950-A

Abstract

The application provides an identification method, a system, electronic equipment and a storage medium of an automobile welding spot stud, wherein the identification method of the automobile welding spot stud comprises the steps of acquiring first image data of multiple visual angles of an automobile part; detecting a welding spot target and a stud target on the first image data based on a specified model to generate detection results of all view angles, screening the detection results of all view angles based on a epipolar geometry constraint principle to obtain homologous target point pairs, calculating the homologous target point pairs by using a triangulation algorithm to obtain candidate three-dimensional space coordinates of the target, carrying out reprojection error verification based on the candidate three-dimensional space coordinates of the target to determine a three-dimensional detection result, and comparing the three-dimensional detection result with preset standard technological parameters to obtain a recognition result.

Inventors

WU FEIBIAO
ZHANG SHAOHUA
FENG QICHAO

Assignees

广州东焊智能装备有限公司

Dates

Publication Date: 20260508
Application Date: 20260128

Claims (10)

1. The method for identifying the automobile welding spot stud is characterized by comprising the following steps of: Acquiring first image data of multiple view angles of an automobile accessory; Detecting a welding spot target and a stud target on the first image data based on a specified model, and generating detection results of all visual angles; screening detection results of all visual angles based on epipolar geometry constraint principle to obtain a homologous target point pair; Calculating the homologous target point pair by using a triangulation algorithm to obtain candidate three-dimensional space coordinates of the target; Carrying out reprojection error verification based on the candidate three-dimensional space coordinates of the target, and determining a three-dimensional detection result; and comparing the three-dimensional detection result with preset standard technological parameters to obtain an identification result.
2. The method of claim 1, wherein the acquiring the first image data of the multiple perspectives of the automotive part comprises: shooting the automobile parts from different preset angles by utilizing a plurality of image acquisition units to acquire a plurality of paths of original images; Performing geometric distortion correction on the multipath original images based on camera internal parameters and distortion coefficients calibrated in advance to obtain a corrected image sequence with multiple visual angles; dividing the corrected image sequence into a plurality of mutually non-overlapping rectangular sub-blocks to obtain a histogram of each sub-block; clipping and limiting the histogram, and evenly distributing the pixel number exceeding a preset threshold value to other gray levels of the histogram to obtain a processed sub-block; and carrying out bilinear interpolation fusion on the processed sub-blocks to obtain multi-view first image data of the automobile accessory.
3. The method according to claim 2, wherein the method for constructing the specified model comprises: acquiring an initial model and a training set; Embedding a channel attention module and a spatial attention module in a trunk feature extraction network of the initial model to obtain a first model, wherein the channel attention module is used for weighting each channel of the feature map, and the spatial attention module is used for weighting the spatial position of the feature map; Setting a shallow high-resolution characteristic branch for detecting a tiny target in a characteristic fusion network of the first model to obtain a second model, wherein the shallow high-resolution characteristic branch fuses deep semantic information and shallow texture information, and the shallow high-resolution characteristic branch is used for detecting a welding spot or stud target with a pixel size smaller than a preset threshold; obtaining a prediction boundary box and a category prediction probability according to the second model and the training set; determining a real boundary box according to the training set; training the second model with a multi-tasking joint loss function, the multi-tasking joint loss function comprising a localization loss section; Calculating the ratio of the overlapping area and the union area between the prediction boundary frame and the real boundary frame to obtain the intersection ratio; Calculating the Euclidean distance between the central point of the prediction boundary frame and the central point of the real boundary frame, obtaining the diagonal distance of the minimum closure area capable of simultaneously containing the two boundary frames, dividing the Euclidean distance by the diagonal distance for normalization processing, and obtaining a central point distance punishment item; Calculating the difference degree between the aspect ratio of the prediction boundary box and the aspect ratio of the real boundary box to obtain an aspect ratio penalty term; subtracting the center point distance penalty term and the length-width ratio penalty term by using the intersection ratio to obtain a positioning loss value of the sample; constructing a dynamic modulation factor based on the category prediction probability, and performing product operation on the dynamic modulation factor and a standard binary cross entropy loss function term to obtain a category loss part; and optimizing the second model according to the positioning loss value, the classification loss part and the training set to obtain a specified model.
4. The method of claim 3, wherein the detection results of each view angle include a detection result of a first view angle and a detection result of a second view angle, and wherein the screening the detection results of each view angle based on the epipolar geometry constraint principle to obtain the pair of homologous target points includes: obtaining a basic matrix according to the relative position relation of the acquisition units; taking the pixel coordinates of the central point of the specified target in the detection result of the first visual angle as input, and performing matrix multiplication operation by utilizing the basic matrix to obtain an polar equation corresponding to the detection result of the central point of the specified target in the detection result of the first visual angle; traversing the center point pixel coordinates of all candidate targets in the same category group in the detection result of the first view angle to obtain a center point pixel coordinate set of the candidate targets; the vertical pixel distance from the center point of each candidate object to the corresponding line in the center point pixel coordinate set of the candidate object is obtained; And if the specified vertical pixel distance is smaller than a preset tolerance threshold, judging that the candidate target in the second view angle and the target in the first view angle form a candidate homologous point pair, and obtaining a homologous target point pair.
5. The method of claim 4, wherein calculating the pair of homologous target points using a triangulation algorithm to obtain candidate three-dimensional spatial coordinates of the target comprises: constructing a linear equation set comprising a first view projection matrix and a second view projection matrix; substituting the pixel coordinates meeting the homologous target point pair into the linear equation set, and solving the linear equation set by utilizing a least square method or a singular value decomposition method to obtain candidate three-dimensional space coordinates of the target.
6. The method of claim 5, wherein the detection result includes position coordinates of an object, wherein the performing the re-projection error verification based on candidate three-dimensional space coordinates of the object, and wherein determining the three-dimensional detection result includes: Back projecting the candidate three-dimensional space coordinates of the target to respective image planes through a first view projection matrix and a second view projection matrix respectively to obtain re-projection pixel coordinates; Calculating Euclidean distance between the pixel coordinates of the re-projection and the position coordinates of the target to obtain a re-projection error; under the condition that the reprojection error is smaller than a preset pixel threshold value, determining that the candidate three-dimensional space coordinates of the corresponding target are true, and obtaining a three-dimensional detection result; and under the condition that the reprojection error is larger than or equal to a preset pixel threshold value, judging that the candidate three-dimensional space coordinate of the corresponding target is false, and eliminating the candidate three-dimensional space coordinate of the corresponding target.
7. The method of claim 6, wherein the recognition result includes a first recognition result and a second recognition result, and the comparing the three-dimensional detection result with a preset standard process parameter to obtain the recognition result includes: Calculating the space distance deviation between the three-dimensional detection result and the standard design coordinate to obtain a target space distance deviation; Obtaining a first recognition result when the target space distance deviation is within an allowable tolerance range; and obtaining a second recognition result when the target space distance deviation is not within the allowable tolerance range.
8. An identification system for an automotive weld stud, comprising: The first acquisition module acquires first image data of multiple view angles of the automobile part; The first generation module is used for detecting a welding spot target and a stud target on the first image data based on the specified model, and generating detection results of all visual angles, wherein the detection results comprise position coordinates of the target, target category and confidence information; the first obtaining module is used for screening detection results of all visual angles based on the epipolar geometry constraint principle to obtain a homologous target point pair; The second obtaining module is used for calculating the homologous target point pairs by using a triangulation algorithm to obtain candidate three-dimensional space coordinates of the target; The first determining module is used for carrying out reprojection error verification based on the candidate three-dimensional space coordinates of the target, and determining a three-dimensional detection result; and the third obtaining module is used for comparing the three-dimensional detection result with preset standard technological parameters to obtain an identification result.
9. An electronic device, comprising: At least one processor, and A memory communicatively coupled to the at least one processor, wherein, The memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.
10. A computer readable storage medium having stored therein computer instructions which, when executed by a processor, implement the method of any of claims 1-7.

Description

Identification method and system for automobile welding spot stud, electronic equipment and storage medium Technical Field The application relates to the technical field of automobile welding spot stud recognition, in particular to a method, a system, electronic equipment and a storage medium for recognizing an automobile welding spot stud. Background In the automobile manufacturing and after-sales quality inspection stage, quality assessment of automobile parts (such as welding spots, studs and other key connecting parts) is a core link for ensuring the safety and the use reliability of the vehicle structure. Currently, quality evaluation of common targets such as automobile welding spots, studs and the like in the industry mainly depends on two modes, namely manual visual detection, namely detection personnel observe whether the targets exist or not by naked eyes, appearance forms and the like, judge whether the targets accord with quality standards by combining experience of the detection personnel, and adopt a traditional machine visual detection scheme, namely unfolding and analyzing the targets by a preset fixed image feature extraction algorithm, so as to judge whether the targets exist or not. In the existing automobile welding spot stud recognition process, analysis and recognition are carried out through a machine vision detection scheme, and although the efficiency is high, misjudgment is easy to occur under the conditions of partial shielding, dirt on parts and the like. Disclosure of Invention The embodiment of the application provides a method, a system, electronic equipment and a storage medium for identifying an automobile welding spot stud, which are used for solving the problems of the related technology, and the technical scheme is as follows: in a first aspect, an embodiment of the present application provides a method for identifying an automobile welding spot stud, including: Acquiring first image data of multiple view angles of an automobile accessory; Detecting a welding spot target and a stud target on the first image data based on the specified model, and generating detection results of all visual angles; Screening detection results of all visual angles based on epipolar geometry constraint principle to obtain a homologous target point pair; Calculating the homologous target point pair by using a triangulation algorithm to obtain candidate three-dimensional space coordinates of the target; carrying out reprojection error verification based on candidate three-dimensional space coordinates of the target, and determining a three-dimensional detection result; and comparing the three-dimensional detection result with preset standard technological parameters to obtain a recognition result. In a second aspect, an embodiment of the present application provides an identification system for an automobile welding spot stud, including: The first acquisition module acquires first image data of multiple view angles of the automobile part; The first generation module is used for detecting a welding spot target and a stud target on the first image data based on the specified model, so that detection results of all visual angles are generated, wherein the detection results comprise position coordinates of the target, the target category and confidence information; the first obtaining module is used for screening detection results of all visual angles based on the epipolar geometry constraint principle to obtain a homologous target point pair; The second obtaining module is used for calculating the homologous target point pairs by using a triangulation algorithm to obtain candidate three-dimensional space coordinates of the target; the first determining module is used for carrying out reprojection error verification based on candidate three-dimensional space coordinates of the target and determining a three-dimensional detection result; and the third obtaining module is used for comparing the three-dimensional detection result with preset standard technological parameters to obtain an identification result. In a third aspect, an embodiment of the present application provides an electronic device, including at least one processor, and a memory communicatively connected to the at least one processor, where the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method for identifying an automotive solder joint stud. In a fourth aspect, embodiments of the present application provide a computer-readable storage medium storing computer instructions that, when executed on a computer, perform a method according to any one of the above-described embodiments. The advantages or beneficial effects in the technical scheme at least comprise: The method for identifying the automobile welding spot stud comprises the steps of obtaining first image data of multiple view angles of an automobile fitting, detecting a welding spot target and a stud targ