Distance Estimation by Fusing Radar and Monocular Camera with Kalman Filter

The major contribution of this paper is to propose a low-cost accurate distance estimation approach. It can potentially be used in driver modelling, accident avoidance and autonomous driving. Based on MATLAB and Python, sensory data from a Continental radar and a monocular dashcam were fused using a Kalman filter. Both sensors were mounted on a Volkswagen Sharan, performing repeated driving on a same route. The established system consists of three components, radar data processing, camera data processing and data fusion using Kalman filter. For radar data processing, raw radar measurements were directly collected from a data logger and analyzed using a Python program. Valid data were extracted and time stamped for further use. Meanwhile, a Nextbase monocular dashcam was used to record corresponding traffic scenarios. In order to measure headway distance from these videos, object depicting the leading vehicle was first located in each frame. Afterwards, the corresponding vanishing point was also detected and used to automatically compute the camera posture, which is to minimize the interference caused by camera vibration. The headway distance can hence be obtained by assuming the leading and host vehicles were in the same ground plane. After both sensory data were obtained, they were synthesized and fused using Kalman filter, to generate a better estimation of headway distance. The performances of both sensors were assessed individually and the correlation between their measurements was evaluated by replotting radar measurements on the video stream. The results of individual sensors and Kalman filter were compared to investigate the optimization performance of the data fusion approach.This is a general guidance of headway distance estimation with a low cost radar and a monocular camera. With described general procedures, this paper can allow researchers to easily fuse radar and camera measurements to obtain optimized headway distance estimation. This paper can facilitate the development of a more realistic robotic driver that can mimic human driver behaviors.