Research On The Lane Detection Algorithm For Lane Departure Warning System

Lane detection is that the continuous or discontinuous lanes in the driving region are reconstructed two continuous curves from the images which contain lane marks. Lane detection is the premise and basis of the whole Lane Departure Warning System. Whether the lanes in the road image can be identified quickly and accurately or not, it is important for the stability and the accuracy of the warning system. In order to solve the problem of poor robustness and low real-time of lane detection algorithm in the condition of unstructured road, illumination variance, shady lane mark, worn lane mark, stain covered lane mark, etc, feature extraction algorithms which have good adaptability and feature points fitting algorithms are proposed.The preprocessing of road image is put forward in order to extract the feature points efficiently, including graying, filtering and dividing the region of interest, fristly. Further more, the feature extraction algorithm of symmetrical local threshold segmentation is proposed,because the classic algorithm can hardly calculate a proper threshold in the view of the whole road image in the complex road condition. Meanwhile, another one is proposed based on the lane mark local structure feature rather than the gray value of the lane mark, that is, the feature extraction algorithm based on Level-set Extrinsic Curvature. Finally, improved RANSAC fitting algorithm is introduced in terms of the feature points which contain not only the good ones,but also many bad points and their distribution is irregular.The validity of proposed algorithms are verified by many experiments in various kinds of complex road images. The results show that, the algorithm of symmetrical local threshold segmentation can extract the feature points efficiently when the contrast between lane mark and road is low. At the same time, another feature extraction algorithm is based on the lane mark local structure. That is to say, it has little relation with the contrast between the lane mark and road, so it can work efficiently in the complex road condition. Improved RANSAC algorithm can calculate the lane parameters whether the lane is curve or straight in the complex road condition and has good robustness. Meanwhile, the iteration process is independent nearly because of deleting the feature points step by step, and the best lane parameters can be more possible to acquire while the real-time is improved. Undoubtedly, this will lay a theoretical foundation for putting lane detection algorithm into practical application which has good real-time and accuracy.