SIGNAL+DRAHT | Issue 09/2024
The visual detection of obstacles in the track bed with scalable sensitivity for ATO
The visual detection of obstacles on the track requires reliable detection at high speeds over large distances, whereby unknown obstacles must be detected. Data-driven learning techniques do not require any specification of the type of processing that occurs automatically, but do require examples of any possible obstacles so as to be able to identify them. Our approach, on the other hand, relies on model-based processing, which in the case of a railway application works with a simple model of an open, planar railway line and visually monitors any deviations without knowing the obstacles in advance...