Machine Vision-Based Straightness Measurement of Pressed Joints

Abstract

To address the issues of complexity, time consumption, and low detection accuracy in traditional straightness detection methods for pressed joints, a non-contact measurement method based on machine vision is proposed. The main approach is to rotate the joint and acquire images at different angles. Bilateral filtering is used to denoise the images. Considering the potential reflections and simple background of the joint, color space conversion and local binary patterns are introduced to capture local texture features. Through feature stitching and principal component analysis to adapt the Grab Cut algorithm, the joint region is segmented. The weighted average method is used to grayscale the extracted image, and after Canny edge detection, edge connection, and filling algorithm, the edge contour image of the joint is obtained. According to the contour information and the straightness measurement method proposed in this paper, the center coordinates of each cross-section of the joint are obtained, and then the least squares method is used to evaluate the straightness error of the joint to judge the bending condition of the joint. Finally, comparative experiments show that the detection system has high detection accuracy and reliable detection results.