Experimental Study on Vehicle to Road Tracking Algorithm by Using Kalman Filter Associated with Vehicle Lateral Dynamics

This paper presents a vehicle to road tracking algorithm based on vision sensor by using Extended Kalman Filter (EKF) from which outputs [i.e. lateral offset, heading angle relative to lane, road width, and road curvature, so called, VRTP (Vehicle to Road Tracking Parameter)] might be used as inputs to steering controller of lane keeping assist system or for smart warning decision logic of lane departure warning system among automotive driver assistance systems. The proposed approach makes use of lane marking pixel coordinates on image plane extracted from a kind of lane detection algorithm, together with yaw rate, steering angle and velocity measurements. The proposed algorithm bears consistent exactness of VRTM even when the camera tilt angle is near zero while the previous works based on Kalman filter frequently adopted shows the degradation of performance because the nearly camera tilt angle is zero, the more the variations of the raw points of lane marker far away from vehicle results in disparity between physical coordinates and image coordinates. To overcome this defect, state evolution model for VRTM is replaced by vehicle to road kinematics considering side-slip angle instead of random walk model. Presented algorithm has been implemented and tested at proving ground. The results have been also compared with random walk model based Kalman filter algorithm by using DGPS-RTK equipment to evaluate the exactness of VRTM quantitatively.