SINGLE TREE DELINEATION USING AIRBORNE LIDAR DATA

Abstract

In this paper, single tree extraction was carried out using the first and last pulse airborne LIght Detection And Ranging (LIDAR) data. The LIDAR data was collected from TopoSys in May 2007 in the Milicz forest district, Poland, with a density of 7 points m-2. The total study area contains 25 circular plots of different radius according to the age of the trees. The absolute height of each point was obtained by normalizing the LIDAR raw data points using a digital terrain model (DTM) of the area. The value of σ used while smoothing was found higher for the deciduous tree dominating plots as compared to the coniferous plots. A modified k-means clustering algorithm was applied to extract the clusters of single tree above 4m height in each plot from the normalized LIDAR point clouds. 3-D convex polytope reconstruction from the extracted clusters of each tree was carried out using QHull algorithm. The validated result shows that an average of nearly 86% of the matured deciduous and 93% of the matured coniferous trees were extracted by the presented approach. Almost equal average accuracies were obtained in the case of young deciduous and coniferous tree species (58%). It seems that the algorithm did not work well with relatively younger tree types even after varying the parameters at pre-processing steps. The study showed that the adjustment of certain parameters like threshold distance, smoothing factor and scaling factor for the height before initialising the main process, has a substantial impact on the number and shape of the trees to be extracted more appropriately by applying the modified k-means procedure. There is a future scope of improving and testing the algorithm with different density of LIDAR data in different forest conditions.