Method for extracting light stripes of wheel profiles from consecutive frames in complex background

This paper proposed a method that involves detecting the enclosing rectangle of binary images in the R component of acquired images, utilizing the shape characteristics of thin and long light stripes and their distinct color characteristics, against complex backgrounds caused by varying illumination conditions in wheel profile detection. By setting filter conditions for rectangular areas, light stripe zones were quickly segmented within complex backgrounds, eliminating the need to repeatedly input threshold parameters, thus enhancing the algorithm’s adaptability and reliability. Additionally, an algorithm of edge expansion in consecutive frames utilized the similarity between image frames to segment light stripes, contributing to the algorithm’s robustness against texture changes on object surface. While maintaining stability, the proposed method achieved an image detection speed 16 times higher than that of single-frame processing methods, meeting the real-time requirements for wheelset profile detection, with an average processing time per frame of 0.22 s.