Probabilistic Distance Estimation for Vehicle Tracking Application in Monocular Vision

Measuring absolute distances in monocular vision is challenging and cannot be solved directly. Conventionally, assumptions like an a priori width of the target object or geometric constraints are made to overcome the problem. In this paper we describe a probabilistic solution that integrates distance estimation in a vehicle tracking environment. This is obtained by using a ground plane angle based estimation together with a width interval constraint utilizing a vehicle classifier. Furthermore the information from a lane departure warning system is fused. We combine width and angular information together utilizing a Bayes estimator. For testing a large video data set with distance ground truth obtained from a radar has been generated. We show that using this estimator enhances the distance estimation in a Kalman filter based vehicle tracker environment compared to the standard constraints widely used. The presented probabilistic integration is very time efficient and has been successfully tested online.