Keywords: LiDAR, remote sensing, railway maintenance, artificial intelligence, point cloud, deposit detection.

A railway scene involves a wide range of components including infrastructures (e.g. rails, catenary poles, ballast, etc.), environmental elements (vegetation, terrain, etc.) as well as temporary objects (stationary trains, machinery, etc.). While most of these components are either meant to be there or do not directly affect railway operations, some objects ideally should not be present on the tracks and their surroundings. These include moderately large objects like rail segments or ballast big bags, or smaller ones such as steel fishplates, nuts, and bolts which typically remain on-site following maintenance operations. While their impact may be limited, they still represent unused material that could be recycled, and can further raise safety concerns if too close to the tracks. In this paper we present an automated LiDAR-based method capable of detecting such object deposits, with a focus on the aforementioned larger elements. The precise modelization and geolocation capacities of LIDAR technology allows us to precisely locate the deposits across the entire railway network in France, and provide measurement estimates of their characteristics. The process relies on a comparison between a reference point cloud of a given railway scene in the past, and a more recent target point cloud of the same scene in which we aim to detect the deposits. The target is aligned with the reference, and any differences between the two point-clouds are tested as actual deposits or not, after filtering irrelevant sources of change such as vegetation.

download the full-text of this paper (PDF, 1685 Kb)

go to the previous paper
go to the next paper
return to the table of contents
return to the volume description