A 3D Visual Detection Model for Bogie Pose of Maintenance Robot Based on Deep Learning

The bogie of freight railcars is a critical component affecting train safety and maintenance automation. Traditional pose localization and bolster tilt detection methods rely heavily on manual measurements and laser ranging, which suffer from high environmental sensitivity, limited accuracy, and poor real‐time performance. To address these limitations, this study proposes a visual pose detection system for bogie bolster tilt based on deep information fusion. The system includes hardware design using depth cameras, spatial alignment of RGB and depth data, and temporal filtering for depth smoothing. An adaptive feature selection module and coordinate attention (CA) mechanism are incorporated into the detection network to enhance feature extraction. Experimental results show that the system achieves pose detection errors within ±4 mm for depth, ±3 mm for positional offset, and ±0.5° for rotation. Further improvements using adaptive filtering, optimized depth map matching, and compensation function refinement reduce depth and offset errors to within ±2 mm, while maintaining rotation error within ±0.5°. The results validate the proposed method’s effectiveness and its potential for intelligent bogie inspection applications.