Study On Algorithms Of Vision-based Vehicle Lane Departure Warning System

Based on analyzing the domestic and overseas research status of various navigation techniques for intelligent vehicle systems, this thesis proposes a vehicle intelligent assistant safe-driving system for practical use by applying computer vision technology, this system is designed for vehicles on highways, in hope that a kind of automatic safe-driving can be realized.The thesis discusses the work of vision processing system, concentrating on how to obtain the information of road and obstacle using computer vision technology for expert system to make warning decision. The thesis presents the development of the science and technique of Lane Departure Warning System and presents some basic concepts of computer vision at first. And then, the model of monocular vision system is established to obtain the features of vision field and to obtain the mapping connections between image coordinates and road coordinates with correlative theories of perspective projection pinhole imaging model and optical geometry. Furthermore, multi image processing methods are studied to enhance the adaptability and flexibility of vision system and compared to put forward the relative good-effect ones based on specific road soundings. The image processing includes smoothing-filtering approaches, edge enhancing approaches and threshold segmentation approaches to the road image, the algorithms this thesis determines are median filtering, Sobel arithmetic operator and simple iterating method of optimum threshold segmentation methods. In the end, a two-step method based on image processing technique and optimization algorithm is proposed to detect lane in structural road environment. The method uses not only the local characters but also the global characters of the input images to recognize lane more robustly, and divides every image into two portions in which the regression is processed respectively. After the process above, this vision sub-system can follow the road and site the obstacle quickly and accurately in practical use, which setup the basis for the next work of expert system controlling decision. As a result of the all work of the thesis, a decision-making method of lane departure warning system is presented.The algorithms proposed by the thesis are used to process actual lane image, and the simulation results show that the algorithms are reliable and effective.