Abstract: The lane marking is very important for the 3D road scene.Automatic extraction of lane marking in the point cloud scene is of great significance for tasks like the measurement of road width and automatic driving.This paper proposed a space-spactral feature for the extraction of lane marking from mobile LiDAR point cloud.This method fully considers the relationship between color,spatial neighborhood and elevation within lane marking point cloud,which aims to classify the point cloud data and extract lane marking.In order to confirm the effectiveness of this method,a highway scene mobile laser point cloud data with manual labeling is used for the experiments,the training data sets and test data sets are selected for the lane marking extraction experiments.The overall accuracy of road markings extraction is 99.64%,which proves the effectiveness of proposed method.

Key words: mobile laser scanning, point cloud, space-spectral feature, height difference, spatial distribution