Vehicle Tracking with Heading Estimation using a Mono Camera System

Abstract: Advanced driver assistance systems (ADAS) is a popular and evolving area of research and development. By providing assistance to the vehicle drivers, ADAS could significantly reduce the number of traffic accidents since 90 % of all accidentsare caused by the human factor. ADAS with cameras provides a wide field of view and thanks to today’s advanced image processing techniques, lots of informationcan be extracted from the camera image. This thesis proposes a method of estimating the heading of vehicles using a mono camera system. The method consists of an extended Kalman filter with a constant velocity motion model to predict the vehicle’s path, fed by classification measurements from machine learning algorithms together with angular rate measurements. Monte Carlo simulations performed in this thesis show promising results. The results on real-world data indicate that the method used to construct the angular rate measurements must be improved in order to reach the same results as obtained from the simulations. An additional measurement, the vehicle’s corners, is introduced in order to further provide the filter with information. The thesis shows that the mono camera system needs further improvements in order to reach the same level of performance as a stereo camera system.