Vision-based track estimation and turnout detection using recursive estimation

Track selective localization of railway vehicles is a precondition to more efficient logistics, improved security and autonomous driving. Often, satellite based navigation systems are used for localization tasks. However, in many cases, satellite navigation is not available or the sensor information is corrupted. To enhance availability and localization quality new sensors are needed. In this work we propose the usage of a monofocal video camera to improve the localization quality. Our algorithm estimates the track recursively in the camera pictures. The result is used for a turnout detection. Compared to GPS/INS curvature and turnouts can be detected in advance. Our approach uses recursive estimation to track the tracks in the pictures and to estimate the geometry of the tracks. Experimental results show the efficiency of the proposed algorithm.