Abstract: According to the results of feature extraction in vehicle laser point cloud, the requirement of automatic vectorization is put forward. In this paper, a method of edge detection and vectorization based on bi-directional integration is proposed. Experiments results show that:The method is feasible in dealing with practical problems. Outlier filtering is applied to the input feature point cloud to ensure the accuracy of the feature point cloud of the outsourcing box algorithm. Three-dimensional point cloud is projected to the optimal plane according to the bounding box algorithm to provide input for the subsequent integration along all directions. According to the eight neighborhood KD-tree algorithm, the mean neighborhood radius of the sample feature point cloud is calculated. The purpose is to provide the differential elements in the integration domain for the integration in all directions according to the neighborhood radius. According to the provided differential element, we integrate in all directions, and find the optimal solution of the maximum distance from the plane in the integral element. After that, the point cloud index is constructed according to the integration results, and finally the vectorized point cloud of high-precision map is obtained. The result not only ensures the basic features of the feature, but also ensures the absolute accuracy of the point cloud.

Key words: vehicle borne laser point cloud, bounding box algorithm, two-way integration method, eight neighborhood algorithm, outlier filtering