Maintaining, renovation and construction of railway projects require accurate and up-to-date maps of the current rail situation. Obtaining these maps currently is an expensive and time consuming endeavour.
Therefore, a joint project between our research group Ambient Intelligence and Strukton Rail was conceived to automate the process of mapping. To do so, a dedicated measurement train is used which is equipped with a LiDAR. This LiDAR accurately captures the rail environment as a point cloud. Using deep learning a model was trained to detect objects within the point cloud.
As a result, digitalisation of large segments of rail can be done in a short amount of time.
This video shows the results from the PointPillar object detection model. For this proof-of-concept, the following object classes are detected: masts, signals, tension rods, and relay cabinets. The PointPillar implementation from the MMDetection3D framework was used.