Real-time motion based vehicle segmentation in traffic lanes

Abstract

The detection of approaching vehicles in traffic lanes is an essential processing step for a driver assistant or a visual traffic monitoring system. For this task a new motion based approach is presented, which allows processing in real-time without the need of special hardware. Motion estimation was processed along contours and restricted to the observed lane(s). Due to the lane based computation vehicles were segmented by evaluation of the motion direction only. The contour tracking algorithm allows a robust motion estimation and a temporal stabilisation of the motion based segmentation. The capabilities of our approach are demonstrated in two applications: a overtake checker for highways and a visual traffic monitoring system.