Fault Detection of Valve Based on 2D Localization and 3D Pose Solution

We present a detection method based on 2D image localization and 3D pose calculation to solve the fault detection of high-speed train valves. In this paper, the lightweight PP-PicoDet detector is selected to identify and locate valve components, and the 3D pose calculation with spatial position alignment is used to solve the valve state. The novelty of this method adopts the preprocessing method of image object detection and the mapping relationship between image and point cloud. Through experiments, the algorithm has an accuracy of 97% in detecting objects within 0.1s, and solves the valve state within 0. 3s with a deviation of less than 1.5°. The experimental results show that the valve fault detection method combined with PP-PicoDet network and 3D state solution has practical engineering significance.