Segmentation approach for terrestrial point clouds based on the integration of graph theory and region growing

In view of the disadvantage of region growing in the point clouds segmentation, this paper proposes a novel segmentation method for TLS data by integrating graph theory and region growing. This method can be divided into four steps: (1) According to the reflectance value of each laser point, the reflectance image can be created directly from the terrestrial point clouds. (2) The reflectance image will be segmented by the graph theory-based method; (3) The seed points will be selected automatically according to the segmentation result of the reflectance image. The growing condition is the normal vector of the seed point and the intensity information of their neighboring points. Then the point clouds data can be segmented by using region growing method. (4) We combine the segmentation results of the above-mentioned two methods. In order to achieve a satisfying segmentation result, the point clouds are segmented with different segmentation thresholds based on the selected different seed points. The point clouds of terrestrial laser scanner FARO LS 880 was used in the experiment to verify the proposed method.