Tree detection in point clouds derived from terrestrial laser scanning

Terrestrial LIDAR measurements in forest stands is often used to gather data for 3D tree models. However, such models require the detection of points representing trees in the scanning field. The present study offers a method for tree detection from a 3D point cloud of forest plantations. Initially the spatial distribution of trees is reconstructed by applying a segmentation algorithm in a transverse slice (1 meter) through the point cloud. This is followed by an algorithm for detecting tree position based on plantation stand row alignment. Finally, the results are presented for validation by the point cloud user. The methods were evaluated over young Eucalyptus spp. stands (i.e. 2, 4 and 5 years) exploring single and multiple positioning of the TLS device inside the circular plots. Results suggest that several TLS stations should be used to reduce shading effects in mapping circular plots. Employing the tree detection method with the visual analysis of point clouds of each plot were identified 100 % of the trees.

Keywords:
forest plantations; single and multiple scan; circular plots; connect components; Eucalyptus spp.