Researches On Lane Detection And Tracking Algorithm Based On Image

With the rapid development of national economy and the expanding scale of city,the number of vehicles is growing, and the issue of driving safety is faced with severechallenges. Through the analysis of a large number of traffic accidents, rear-endcollision is the accident that cause heaviest casualties and economic losses，while70%of rear-end collision accident is caused by the changing lanes. Automotive driverassistance systems as the guarantee of traffic safety will remind the driver when thevehicle left the driveway or impending collision. Lane detection and recognition as asignificant part of the driver assistant system, including lane departure warningsystem and lane keeping assistance system, has important research significance andvalue. Therefore, how to accurately and effectively detect and track lanes is becominga primary issue of driver assistant system. This article will focus on a study of lanedetection and tracking algorithm based on image.In this paper, we introduce the research status of lane detection and trackingtechnology. Based on the investigation of exising methods on lane detection andtracking, an effective lane detection and tracking algorithm is able to adapt toillumination change is proposed in this paper. The main contents of this paper are asfollows:1. In lane information extraction, we analyze the commonly used lane markingdetection and tracking algorithm, and propose a lane extraction method based on localfeature information. First, the local median filter is utilized to convert the grayscaleimage to a filtered image without bright and dark highlighted objects. Then accordingto the orientation of each connected region, the interference of non-lane markingregion is eliminated and the accurate lane markings are screened out. This algorithmhas strong adaptability and able to adapt to low-contrast situation.2. In the aspect of lane detection, we first introduce the method of fitting straightlines and curvesroad models. In order to make the result closer to the real shape of thelane, and considering the complexity of the algorithm computing, a combination ofstraight line and curve line model is proposed in this paper. Hough transform withsome subsidiary conditions is used to detect straight line and Random SampleConsensus is employed to fit third degree B-Spline model. The proposed algorithmused curve model is able to obtain better results in the case of straight line model is not enough to fit the curve lane.3. In the aspect of lane tracking, we firstly analyze the tracking algorithm basedon the model parameters-particle filter, and give a brief description for video imagetracking CONDENSATION algorithm. Then an algorithm of lane tracking based onCONDENSATION particle filter algorithm is proposed in this paper. By predictingand updating the lane model parameters the lane state is estimated and the purposeoflane tracking is achieved. This method can effectively improve the accuracy androbustness of detection results.The lane marking detection and tracking algorithm evaluated in different imagesequences, including fluctuating illumination and various road scenes, and theexperimental results demonstrate its effectiveness and robustness.