Abstract: The acquisition of detailed information for the streetlights in a large scene remains a tough task since the streetlights are of great number and types. In this paper, a method is proposed to extract and classify the streetlights, with the aid of prior sample sets on the basis of skeleton-line-buffer discriminant algorithm. First, a model and a priori sample set for streetlights are established according to the expression characteristics of streetlights in vehicle-borne LiDAR point cloud. Secondly, with the theory and method of mathematical morphology, the rod-shaped objects are extracted in vehicle LiDAR point cloud scene, and the candidate streetlights are chosen under the constraint of streetlight model and semantic rules. Then, the candidate samples are selected from the sample sets according to the parameter information and the statistical information obtained from the selected streetlights. Finally, based on the matching algorithm of least squares theory, we select and match the priori samples of streetlights and the candidate streetlights. Based on the double buffer of streetlight skeleton information, we discriminate and analyze the candidate streetlights to achieve the extraction and identification of street lights. Finally, the priori samples of streetlights and the point cloud of the candidate streetlights are matched and screened with the matching algorithm of least square theory; and based on the double buffer of streetlight skeleton information, the candidate streetlights are discriminated and analyzed to achieve the extraction and identification of streetlights. Our experiment shows that the algorithm is efficient and robust for the extraction of detailed information of streetlights. For the streetlights with less occlusion and relatively complete data, the extraction accuracy is 0.952, and for those with serous occlusion, low point cloud density and poor data integrity, the extraction accuracy is 0.780. And the above results validate the robustness of the proposed algorithm for the extraction of intermediate streetlights from large scenes. The detailed information extracted by the algorithm can be used to serve the fine and dynamic management of streetlights in large scenes.

Key words: mobile LiDAR, streetlight model, streetlight priori sample, point cloud matching, type recognition