Research On Lane Departure And Collision Warning Based On Machine Vision

In recent years,with the development of economy and society,traffic accidents caused by lane departure and vehicle collision are also increasing.In order to reduce the number of casualties and property damage caused by these accidents,the major automobile manufacturers and research institutions have increased the research investment of vehicle safety warning system.Among them,the security warning system based on visual sensor is favored by domestic and foreign researchers because of its abundant information and low cost.Based on the BJUT-IV intelligent vehicle platform of Beijing University of Technology and visual sensor,the lane departure and collision warning technology based on vision are studied in the dissertation,which include coordinate transformation between image coordinate system and geodetic coordinate system,lane detection,vehicle detection,lane departure and collision warning algorithm.The main work of this dissertation is as follows:1)The hardware and software system of the BJUT-IV intelligent vehicle platform and the selection of the visual sensor are introduced.And the coordinate transformation between the image coordinate system and the geodetic coordinate system is studied according to the imaging model of the visual sensor.2)In order to solve the problem that the threshold segmentation algorithm in the process of lane mark detection is difficult to obtain good results in the case of vehicle,tree shadow,road fence and pavement whitening,a lane detection algorithm based on improved SIS threshold segmentation is proposed.Firstly,according to the gray level difference between the lane and the background,an improved SIS threshold segmentation algorithm is proposed.Compared with the common threshold segmentation algorithms,such as iterative method,maximum entropy threshold segmentation method,Otsu method and Bernsen method,the proposed algorithm has a better adaptability in complex environment.Secondly,the LSD algorithm is used in the straight line extraction phase,which is less time-consuming than the Hough transform method.Then the parallel line pairs are sought in the straight line of the output to estimate the vanishing point position,and then the vanishing point is used to remove the false detection.Finally,use the continuity of the lane line and the lane distance to establish the lane area of interest.And then,in the area of interest,we can find the straight line that is closest to the front frame lane.Compared with the traditional model-based algorithm,this method can automatically adjust the number of interest area of lanes according to the number of lanes.3)A vehicle detection method based on the shadow of the vehicle and the Haar feature is proposed.In order to solve the problem that the Haar feature and cascade classifier can not meet the real-time requirements,the shadow region is segmented by the gray difference between the shadow of the vehicle and the surrounding environment.And then the image is processed by morphology.Finally,the outer rectangle which can contain all the shadow regions is treated as the region of interest,and the Haar feature and the cascade classifiers are used to detect the vehicle in the region of interest.The experimental results show that the proposed method can improve the real-time performance of the algorithm without reducing the recall rate and accuracy rate.4)A method of estimating the position of the vehicle based on the position of the lane line in the earth coordinate system is proposed,and the lane departure warning is provided for the vehicle.Then the calculation method of the distance between the vehicle and the front vehicle in the earth coordinate system is studied,and the anti-collision warning is provided for the vehicle according to the distance information.Finally,the test software based on MFC is designed to verify the practicability of the algorithm.