Thickness Measurement of Brake Pads Based on Keypoint and Object Detection

• Published in: 2023 IEEE International Conference on Unmanned Systems (ICUS) pp. 104 - 109
• Main Authors: Chang, Yujia, Li, Ran, Ma, Dengxin, Ma, Xiangyu, Song, Dalei
• Format: Conference Proceeding
• Language: English
• Published: IEEE 13.10.2023
• Subjects: Brake pad thickness measurement; Current measurement; deep learning; edge detection; Feature extraction; Heating systems; Image edge detection; keypoint detection; Lighting; Location awareness; object detection; Position measurement
• ISSN: 2771-7372
• DOI: 10.1109/ICUS58632.2023.10318238

The wear condition of the brake pad, as one of the core components of high-speed trains, needs to be detected and identified based on its thickness. However, existing measurement methods primarily rely on direct detection of the edges of brake pads, which is sensitive to lighting and contrast conditions. As a result, the accuracy of the algorithm for applications is limited. In this paper, we propose a brake pad thickness measurement method based on keypoint and object detection. We apply YOLOV5 for brake pad localization and adopt the HRNet network for keypoint detection in order to calculate thickness. The positions of keypoints are adjusted based on detected edge features. Precise distance is achieved by utilizing depth infor-mation and world coordinates of keypoints to obtain the actual thickness of the brake pad. Experimental results demonstrate that compared with the original method, our proposed algorithm achieves an average error rate of 0.53mm, which is accurate enough for real-world applications using our self-made dataset in realistic environments.