High- precision extraction of urban road markings based on 3D laser point cloud
HONG Jianwu;Fujian Geological Surveying and Mapping Institute;
In automatic navigation and driving, high-precision map rendering is essential. However, there exist a problem of accuracy loss for the most widely used image recognition technology in this field. Therefore, it is very necessary to use vehicle mounted 3 D laser point cloud for high-precision map rendering. This paper collects point cloud data from the most common urban environment. Firstly, the road boundary is determined by road boundary detection, and combined with a new grid index method, the extraction of road point cloud is realized. On this basis, this part of pavement data is used for automatic extraction of road markings, and the point cloud data of road markings are preliminarily screened by reflection intensity. Secondly, the clustering feature method is used to extract the point cloud data of road markings. Through the measurement of lane width, flatness and segmentation, various lanes are measured. Finally, a high-precision lane model is established, and the samples are tested. The result shows that the integrity rate of the lane recognition can reach 95.3% in this paper, and the minimum deviation degree is only 2%. Meanwhile, the optimal deviation distance can be reduced to 2.3 cm, which can meet the requirements of high-precision rendering.