Research on Train Localization Method Enhanced by LiDAR and Visual Geometric Constraints

This paper proposes a multi-sensor fusion localization method for rail transit, integrating IMU, LiDAR, and visual data with geometric constraints derived from track structures to enhance stability and accuracy. Pose estimation is initialized via IMU-odometer pre-integration, followed by point cloud de-distortion and feature extraction. Track planes constrain pitch and roll angles, while vanishing point detection and perspective correction improve yaw observability. To balance accuracy and realtime performance under high-speed conditions, the system employs region-based point cloud processing and sliding window optimization. Experiments show that the method achieves robust, continuous localization in environments with repetitive structures, high-speed motion, and weak textures, meeting the demands of autonomous train positioning.