Research Lane Detection Technology Vision

The technology of Vision-based Lane Detection is the basic of the drivingsafety assistance and automatic drive; it is also a hot research field of intelligentvehicle at home and abroad. In this paper, we research on the technologies ofvisual-based lane detection in highway environments. Theoretical analysis andsimulation experiments show lane recognition algorithm is robustness, real-timeand efficient.By contrast comparative analysis, we present an improved method of imagepre-processing, which can effective remove useless noise, but also highlight lanefeature information in road image. The method includes: graying, improvedmedian filtering, and contrast enhancement.In order to overcome disadvantages like slow search speed and highrandomness of seed points selection, a region growing method based ondiamond search is proposed. Firstly, light and stains factors which influence thequality of road image and on-board projection principle are studied. Regiongrowing method is improved by using prior knowledge and diamond searchideas to greatly raise speed and accuracy in segmentation processing. Thenimproved method is used to segment the lane boundary effectively. In the end,points set of lane feature have been accurate obtained using fused edge data ofsobel detectors.Finally, the road model is simplified by selecting the line model and theconsistency assumption of road features. To solve the undetected problemscaused by discontinuous lane edge and shadow, straight line is extracted usingmorphological closing operation and probabilistic Hough transform. Thesimulation experiments to lane detection algorithms analyzed and processedroad images sequence on the VC platform with OPENCV, and the simulationresults verify the validity of the theoretical analysis. Experiments show that thelane detection algorithm is put forward to solve effectively issues of anti-lightand anti-shadows, to meet the requirements of real-time and robust in the vehicle system.