Lane Detection On Structured Roads

Lane detection technology is an important means for intelligent vehicle to perceive the environment.Advanced driving assistance systems such as lane departure warning and lane keeping which is based on lane detection technology can effectively reduce the risk of collision.In autonomous vehicles,lane detection technology can indicate the driving area,as an important auxiliary means.The research developments on lane detection are analyzed,the characteristics of lane expressed in images and the difficulties of lane detection are discussed.Aiming at the requirements of real-time,accuracy and robustness of lane detection,combined with the camera model,a lane detection method based on the orientation of vanishing points are proposed in this paper,and the lane model estimating is completed by a combination of straight line and Bezier curve.Specific research content includes the following aspects.1)The detection of the lane straight line is performed in a way that the vanishing point and lane area iterate each other.Combined with the camera model,the relatively complete lane line structure is retained by local feature statistics,and obvious interference features are eliminated.Based on the guidance function of vanishing point,the method of extracting the trusted area of lane line is designed through histogram statistics;and on the basis of the trusted area,the orientation of vanishing point is reestimated.The reliable vanishing point and the trusted area are obtained by the mutual iteration.A cost function is defined to estimate the lane straight line.2)A lane model combining a straight line and a quadratic Bezier curve is established.Two nodes of the quadratic Bezier curve are determined by the straight line.And the estimation of the last node of the Bezier curve is directly completed by particle filtering.In order to ensure the stability of the curve,the last node is associated with the distance from the vanishing points of both sides,and the estimation is completed by a particle filter.At the same time,the real-time performance is also guaranteed.The calculation of particle weight is designed according to the statistical distribution characteristics of lane lines.3)Besides,in order to ensure real-time performance,in operations involving the whole image,such as feature extraction,noise filtering and likelihood calculation,the scan-line for neighborhood averages and statistics can be implemented recursively to reduce calculating time.And coordinate mapping space is constructed based on Bresenham’s line with different width and height to improve the efficiency of the histogram statistical process.4)Finally,experiments were carried out on the open dataset and the dataset of Wuhan University.Experimental results show that the proposed algorithm can meet the needs of lane detection in different scenarios,and has good real-time,accuracy and robustness.