Research On Methods For Vehicles Vision Navigation

Intelligent vehicle technique is a pop research area of ITS (Intelligent Transportation System), and the research on road recognition, which is a key technique for intelligent vehicle navigation system, does great contribution to safe driving and even unmanned driving.In this thesis, to prevent an intelligent vehicle from departing the lane in the vision-based navigation, an integrated method based on monocular vision is proposed to detect the lane marking and road boundary in front of the car. The relative position and direction of the lane and vehicle are then acquired. During the general designing of a vision-based navigation system, the software processing algorithm is presented as below.Firstly, the color-gray transform, image smoothing and contrast enhancement algorithms are applied to the image preprocessing. Through the projection method, the level demarcation line between the road surface and the background is found out, and then the region of interest is defined. Secondly, the characters of the lane marking and road boundary are extracted with the help of the methods like gradient operator convolution, gradient angle quantization, non-maximum suppression and edge point searching. For the particular brightness of the lane marking, an OTSU method is also applied to the lane marking characters extraction. Thirdly, the straight and crooked road models are built according to people's driving experience, prior knowledge and mathematic description. Then the characteristic points of the lane marking and road boundary are matched to the corresponding model by least-squares fit.In this way, the position and shape of the road are identified. What's more, the reference path for driving, the actual position deviation and angle deviation of the driving car are studied in the thesis. A circular processing method is designed for mass image sequences using the results of former frame to reduce the searching range, so the calculating efficiency is increased to a great extent.Most of the algorithms in the paper have been validated by the videos captured from real traffic road scenes, and the experimental results show that the methods are efficient, stable and accurate.