Research On Lane Departure Warning System For Structured Roads

The rapidly developing highway traffic in recent years has become one of the main transportation modes. Traffic accidents occurred more and more frequently at the same time, and people started to pay more attention to road traffic safety. Under such circumstance, vehicle safety driving assistant techniques have become a research hotspot in automotive safety field. Since the driver fails to concentrate on driving, the vehicle deviates from current lane in the running process, which is one of primary causes of traffic accidents. Lane Departure Warning System can detect the vehicle's position in the lane in real time. Once the vehicle leaves the current lane and the turn signal isn't turned on by the driver, the system will warn the driver to correct driving direction in some kinds of methods, such as loudspeaker, indicator light or seat vibration.This paper mainly explores the implementation of lane departure warning technology in structured roads, and designs a system including three kinds of algorithms:road image preprocessing algorithm, lane detection and tracking algorithm, lane departure warning decision algorithm. The main content of this paper is as follows:Firstly, the algorithm of road image preprocessing under complex illumination conditions is discussed. An improved median filtering algorithm is proposed, which can eliminate the road image noise effectively and retain lane edge information. Meanwhile, an adaptive method based on OTSU algorithm, which is used to segment road image, is proposed. Judging from the verification of test, this method shows good segmentation results under four kinds of illumination conditions, including normal illumination, strong illumination, weak illumination and night situation.Secondly, previous lane detection technologies are reviewed in the field of lane detection. In structured roads, the linear lane model is selected. Besides, the probabilistic Hough transform is improved. Tests prove that accuracy and real-time performance of lane detection proposed in this paper have both been improved. Kalman filter is applied to track the lane after detecting the lane correctly. Based on analysis and contrast with previous lane departure warning decision algorithm, this paper proposes an early warning model based on course angle of vehicles, to realize lane departure warning function. Through verification of tests, the accuracy of the model can meet the demands.Finally, the offline simulation test is carried out with road video images captured by camera. Test result indicates that the lane departure warning system in this paper is adaptive to complicated light conditions, meeting the requirements of accuracy and real-time performance, with low false alert rate and low missed alert rate, which satisfies the needs of practical application on road.