Recognition of bus stops through computer vision

This paper introduces a real-time Image Processing System designed for autonomous guidance of passenger buses at bus stops. It continuously estimates the position of the bus relative to a world coordinate system, given by markings, stop lines or other patterns. The pose estimation is based on the monocular image sequence delivered by a forward-looking camera. The system matches the model of an upcoming bus stop against the grabbed images and estimates the 6 degrees of freedoms of the camera. Cinematic constraints are taken into account by means of a Kalman Filter. Since bus stops do not obey a standard geometry, the models are built from the image sequence directly in a teach-in phase. Although the system is proposed for recognition of a Bus stop, it can be applied to recognize any well-described pattern, e.g. exits on highways or marked parking lots.