The Robustness Algorithm Of Lane Departure Warning System Based On Information Fusion

With the rapid growth of Chinese economic and the Motor vehicles’ number, the incidence of traffic accidents are also increasing. The accidents caused by lane departure account for a large proportion. In order to solve this problem, the lane departure warning system is born at the right monent and the research on system robustness is particularly important. In this paper, research work is as follows:1. This study introduced a novel lane detection algorithm(the difference of luminous density method). The proposed algorithm used adapted median filtering and edge detection of the four direction template logarithmic Prewitt based on the difference of luminous density. It gived the threshold of binaryzation. Determined the ROI, so greatly reduced the amount of computation. Lane marking width and temporal trajectory strategy removed false lanes caused by lighting change, and vehicle occlusions. Finally the algorithm got complete and clear lane line. The proposed method is proved to be better than Hough transfom in terms of execution efficiency.2.This paper researshed the lane departure warning system based on information fusion and has considered the relevance and complementarity of each information source. GPS got the position information of vehicle and displays in digital map, then clothoid model extracted the road geometry in front of the vehicle. Lateral kalman filter fused the lane data of camera and map.and finally with vehicle dynamics. In the case of camera can’t get lane imformation.the fusion system improves the stability of road curvature and lane keeping.3. In order to improve the robustness of lane departure warning system, this paper proposed a warning algorithm based on the lane inside boundary point. It reduced the influence of deviation caused by lane detection and lane model distinguish. The calculation formula of Vehicle trajectory distance was deduced from Kinematics of lateral vehicle motion, the algorithm calculated time between two points with the current vehicle speed and steering angle. The algorithm would divide edge points into four classes, then multi-directional search could eliminate the false lane boundary point, and the comparison of the time checked it again. Setted time threshold.the data of Many image frames determined whether deviating from the lane, the algorithm compared with the current mainstream algorithm TLC at last, the experiment shows the advantages of the former.