Research On Lane Detection Based On Binocular Camera

With the frequent occurrence of traffic accidents,the research of intelligent transportation system which facilitates the safe driving of automobiles has been paid more and more attention by scholars.The lane departure warning system is an important field in the research of intelligent transportation systems.In the lane departure warning system,the lane acts as the navigation identifier,it's detection technology has become the focus of research object.Accurate and efficient detection of lane is of great significance to the realization of lane departure warning system and autonomous navigation system.This takes the perspective distortion which is caused by the visual sensor imaging principle as the starting point,due to the changes in camera parameters which is caused by the movement of vehicle,we also consider the changes,and conduct the research of lane detection that is based on the binocular camera.The main research contents are as listed follows:(1)Binocular camera self-calibration: Aimed at the uncertainty of the camera parameters in the process of driving,we studied the methods of binocular camera parameters self-calibration.We utilized the normalized 8-point method which is based on SIFT feature point to solve the fundamental matrix.Meanwhile,we used the improved block sampling LMedS algorithm to carry on the fundamental matrix's robust estimation.Combined the fundamental matrix with the off-line calibration of the internal parameter,we decomposed and optimized these parameters to get binocular camera gesture relationships.(2)Ground-plane inverse perspective projection transformation: At present,the inverse perspective transformation method of monocular camera is fixed and complex for computation,aimed at these disadvantages,we designed a new inverse perspective transformation model which is based on three-dimensional point.Combined with the depth information which is acquired by the binocular camera,we estimated the three-dimensional point on the plane.With the optimizing transformation,we got the high quality images.Our method can effectively improve the applicability and instantaneity of algorithm.(3)Lane detection in transformation view: After eliminating the perspective of deformation,we can acquire the vertical view,we carried on the noise suppression for the images,and extracted the edge information of the lane.Finally,we took advantages of Hough Transform and RANSAC algorithms to detect the lane line,and showed it prominently on the source image.