Abstract: The method of lidar scanning point density is widely applied to lane detection based on lidar, but it has poor anti-interference performance. As an improvement, the echo pulse width of lidar is used to extract features because of its good differentiation between road surface and lane. What's more, a feature extraction method is proposed which contains 2 steps: the seed points extraction based on dynamic threshold of echo pulse width, and the region growth based on Gaussian kernel weighting searching. Then, fuzzy C-means of line (FCL) algorithm clustering based on line character (based on the center of the line) is introduced to identify the lane markings. Finally, least square method is used to fit lane markings. The method is tested on an unmanned vehicle application platform under six different road conditions, and lanes are all detected with high detection precision.