A fusion system for real-time forward collision warning in automobiles

We describe a fusion system that combines vision-data from a single forward-looking camera with forward-looking radar data for real-time forward collision warning in automobiles. Our approach employs computer vision techniques to primarily perform the detection and tracking of vehicles and overhead structure. These detections are then fused using a probabilistic framework with co-registered radar data to reliably obtain vehicle azimuth and depth by minimizing false alarms. The resulting detections can then be used as input to any forward collision warning system. Experimental results are presented to illustrate the performance of the algorithm.