Segmenting Individual Tree from TLS Point Clouds Using Improved DBSCAN

Terrestrial laser scanning (TLS) can provide accurate and detailed three-dimensional (3D) structure information of the forest understory. Segmenting individual trees from disordered, discrete, high-density TLS point clouds is premise for obtaining tree parameters understory, pest control fine modeling. In this study, we propose a bottom-up method to segment data based on density-based spatial clustering applications with noise (DBSCAN). addition, also improve DBSCAN distance distribution matrix (DDM) automatically adaptively determine optimal cluster number corresponding input parameters. Firstly, proposed improved detect trunks obtain initial results. Then, Hough circle fitting used modify trunk detection Finally, segmentation realized regional growth layer-by-layer clustering. paper, use multi-station Chinese artificial German mixed forest, then evaluate efficiency three aspects: overall segmentation, small segmentation. Furthermore, compared existing methods. The results show that total recall, precision, F1-score are 90.84%, 95.38% 0.93, respectively. Compared traditional DBSCAN, accuracy increased by 6.96%, 4.14% 0.06, result comparable those methods, be extracted under tall accurately segmented.