Highway overhead structure detection with on-line camera pitch bias estimation

A radar-based forward collision warning system has the draw back of not being able to distinguish a stopped vehicle from the false alarms caused by highway overhead structures. A vision-based overhead structure detection system has previously been developed that detects and tracks horizontal edge features from the overhead structures. The system also estimates the height of the structures and their distance to the host vehicle, which can be used to help reject false alarms from radar caused by the overhead structures. The algorithm used in this system relies on accurate calibration of the in-vehicle video camera, and the so-called reference horizon in particular, which can vary depending on the camera pitch bias. In this paper, we present an algorithm for on-line estimation of the reference horizon, and consequently the pitch bias of the camera. An enhanced system combining overhead structure detection and reference horizon estimation, and steps towards real-time implementation are also presented.