Multi-lanes detection based on panoramic camera

摘要

The lane detection system is one of the most important subsystems to achieve the environmental perception of autonomous vehicles. This paper addresses the problem of detecting multiple lanes in a relatively large range around the autonomous vehicle without making too much approximation on the shape of the lane marks. A panoramic camera and a LIDAR is used to obtain a wide range of image information and exclude part of the interference information. The vertical lane marks with specific width is extracted by a filter with 2D Gaussian kernel. In order to describe the lanes with relatively complex shapes like variable curvature curves, the model of equidistant curves is proposed and a robust detection method of equidistant curves is designed under the guidance of the lane model. Satisfactory experimental results in diverse environment and the successful application on autonomous vehicles demonstrate the effectiveness of the proposed method.