Track and turnout detection in video-signals using probabilistic spline curves

Track selective localization for rail vehicles is a task that is not yet solved completely as far as autonomous localization methods are concerned. Satellite-based navigation can not fulfill all demands as the signal might not always be available. The localization information might be corrupted as well. So, a fusion with other sensors can be useful to enhance availability. In this paper a monofocal camera based turnout detection is described. It can be used for localization when used together with a digital map. The tracks are represented by spline curves that are adjusted in every frame by a recursive estimation algorithm. The efficiency of the algorithm is shown through experimental results.