Computer vision based train rolling stock examination

Train Rolling stock examination involves visual observation of the moving train around 30Kmph to find defective bogie parts. A train coach moves on a couple of bogies consisting of wheels, suspension and other binding hardware. The health of the bogie decides the safety of the train. Railway personnel perform the rolling stock examination manually raising questions on reliability. Here we propose to use computer vision algorithms for extraction and localizing defective bogie parts from working parts. A wide-angle high-speed camera captures the moving train without motion artefacts. The objective is to use a single shape prior to power the level set function for object segmentation. Here we show the bogie part segmentation with one shape prior model for the entire length of the train. Experimentation on similar train bogies under different lighting tests the robustness of the level set functional with single shape prior. The proposed algorithm handles topological spatial deformations of the bogie parts in the video effectively. Segmenting defective parts with non-defective shape priors makes the algorithm independent of defect localization in the bogie part. This novel idea of computer vision based rolling stock examination using high-speed video can lessen human errors and aid in developing a crewless rolling stock examination. Further, the proposed work can be extended for early detection and prevention of rail accidents due to transit part failures.