Fault Detection and Exclusion for LIDAR-Camera Fusion in Railway Positioning: Enhancing Localization in Adverse Environments

Accurate and reliable environment perception is crucial for the safe operation of future automated train control systems. The objective of this work is to design and validate a two-stage integrity architecture for railway positioning, focusing on Fault Detection and Exclusion (FDE) methods applied to camera and LiDAR data. At the raw-data level, camera integrity is assessed using no-reference image quality metrics (MAD, PIQE, BRISQUE, NIQE) to identify and discard degraded frames, while LiDAR integrity is ensured through statistical, range-based, and density-based outlier removal techniques. At the fusion stage, stereo-camera and LiDAR point clouds are temporally aligned and registered via Iterative Closest Point (ICP), with FDE applied to the merged dataset to exclude anomalous measurements. Experimental results on real railway scenarios, including traffic lights, power line poles, and speed profile markers, demonstrate that the proposed pipeline effectively reduces outliers and enhances the integrity of the fused point cloud, supporting high-accuracy and high-integrity train positioning.