Rail Track Extraction from Rasterized Classified Point Clouds Using a Full-Resolution, Fully Convolutional Recurrent Neural Network

2026-07-07

AI Summary

This study addresses the insufficient accuracy in automatic railway track extraction by proposing an end-to-end method based on a full-resolution fully convolutional recurrent neural network (FCRNN). The approach rasterizes classified point clouds and employs FCRNN for denoising and enhancing track contours, followed by morphological operations, centerline smoothing, and dynamic time warping (DTW) to reconstruct high-precision 3D track centerlines from 2D extraction results. Its novelty lies in being the first to apply FCRNN at full resolution for track extraction and relying solely on synthetic data for training. Experimental results demonstrate that the method automatically generates high-quality rail top surfaces and dual-track centerlines with minimal human intervention, significantly improving pixel-level accuracy and automation.

Abstract

Rail track extraction is essential for effective railway asset management and maintenance, especially in automated inspection and mapping workflows. This paper introduces a novel method for extracting rail tracks from classified 3D point clouds using a fully convolutional recurrent neural network that preserves full spatial resolution and is trained exclusively on synthetically generated data. This approach enhances per-pixel quality and is particularly suited for rail track extraction. The proposed method begins by rasterizing points corresponding to railroad tracks, then applies the neural network to reduce noise and yield a cleaner track representation suitable for vectorization [1]. Subsequent morphological operations further refine the resultant data, enabling accurate track centerline extraction. Next, the extracted centerlines undergo smoothing to eliminate residual irregularities [2, 3]. Finally, the algorithm transfers 3D information from lidar points onto 2D polylines and applies additional vertical smoothing. A single centerline for both tracks is found using the Dynamic Time Warping (DTW) algorithm [4]. The final outcome consists of rail top centerlines and track centerlines derived for rail pairs, with minimal manual intervention. Experimental validation confirms the effectiveness of this method in yielding high-quality rail track extraction.

Problem

• Rail track extraction
• 3D point clouds
• Rasterization
• Centerline extraction
• Railway asset management

Innovation

• fully convolutional recurrent neural network
• rail track extraction
• synthetic training data
• full-resolution processing
• Dynamic Time Warping