Research On Lane Detection And Warning Algorithm Based On Image In Complex Scenes

With the development of social economy,people’s use of cars has become increasingly popular.The application of the vehicle’s driver assistance system is of great significance in reducing the probability of traffic accidents,decreasing the loss of life and property,and ensuring the safety of driving.Due to the similar perception between the vehicle’s and people’s visual sense,most of the driver assistance systems,researched on relative low-cost computer vision technology,are occupied essential positions by the leading lane departure and detection.This thesis takes lane detection and departure warning algorithms as the research target,with combining actual application scenarios to study the lane detection and warning algorithms in complex scenarios which is based on the images.It also focuses on the lane detection algorithm which is for curved lanes and under shado w interference,and the algorithm of the lane departure warning.Aiming at the curve lane detection problem in common complex scenes,a curve lane detection algorithm based on bidirectional edge filtering is proposed.The edge distribution function is used to accumulate the gradient amplitudes of the edges in the image to determine the direction of the straight lane lines in the near field of view and filter out irrelevant edges.Based on the improved Hough transform method,the position and parameters of the near-field straight-line part of the curve lane line in the image are combined with the near-field lane line parameters to extract the characteristic edge points of the lane line of the entire field of view with a bidirectional edge filtering algorithm.According to the extracted edge points,it sets up a model constructed by a straight line and a B-spline curve to fit the curve lane line.Experimental test results show that the proposed algorithm can achieve better detection results than similar algorithms.Aiming at the problem of shadow interference in common complex scenes that influences the effect of lane detection,the lane detection algorithm is studied.A method based on IPM image wavelet decomposition and vertical sub-graph fusion reconstruction is proposed to effectively eliminate the interference of shadows on lane detection.The top view of road image is obtained by inverse perspective transformation.The wavelet decomposition image is used for road top view and vertical sub-graph fusion is used.The Canny edge detection algorithm and the improved Hough transform method are used to detect the feature points of the lane line,and the least square method is used to fit the lane line.A comparative experiment on the removal effect of shadow interference by this method is designed,and experimental results show that the algorithm is robust to shadow interference.In the study of the lane departure warning algorithm,the classic lane departure warning algorithm is analyzed and compared.The straight line equation detected by the lane line is used as a parameter to construct the lane departure warning algorithm based on the model of the lateral offset distance and the heading offset angle of the image,and it’s designed to verify the effectiveness of a set of comparative experiments.Aiming at the early warning problem of high false alarm rate,an iterative algorithm based on lateral offset distance prediction is proposed to predict the future horizontal position,and the judgment of the offset distance is integrated on the basis of the TLC algorithm to improve the early warning judgment conditions.It will occur only when the conditions,which are of the lateral moving distance and the time to approach the lane line,are both met,and that will reduce the numbers of the false alarms.