Fast Extraction Method for Railway Trackside Facility Features

To enable efficient detection of railway trackside facilities and real-time train positioning, this paper presents a fast and lightweight feature extraction method. The approach integrates three-dimensional laser point cloud processing, image-based clustering, geometric constraints, and deep learning for multi-modal detection. Specifically, pole-like features are identified through depth image projection and geometric filtering, intensity-based segmentation extracts facilities with strong reflection signatures, and kilometer posts are detected by a light-weight YOLOv5s model enhanced with the C3Ghost module. Digit recognition on kilometer posts is achieved with a compact three-stage OCR network. Experiments on multi-scenario datasets and embedded hardware show that the method achieves accurate detection while maintaining real-time performance, with kilo-meter post recognition accuracy of 97.4% at 35 FPS. This work provides a reliable perception foundation for intelligent train control systems.