Automatic Rail Detection Technology Based on PointNet++ Using 3D Point Cloud Data of Railway Bridges

Abstract

Recently, railway maintenance has been receiving significant attention to prevent railway accidents. Accordingly, various methods are being developed that apply IT to railroad maintenance, and digital models can be used for an efficient management. To develop a railroad digital model, current status information of the rail is required. However, the existing method consumes considerable time and cost. Therefore, in this study, we proposed a system to scan the railroad using a UAV and automatically detect the rail using PointNet++. The proposed system consisted of Phase 1 (structure from motion) and Phase 2 (rail detection). To verify the performance of the proposed system, the railroad bridge of the Osong test track in Nojang-ri, Jeondong-myeon, Sejong City, South Korea, was targeted. The proposed system is expected to be utilized in various fields such as damage detection, simulation, predictive maintenance, and efficient operation management.