Individual Tree Detection Based on Densities of High Points of High Resolution Airborne Lidar

The retrieval of individual tree location from Airborne LiDAR has focused largely on utilizing canopy height. However, high resolution Airborne LiDAR offers another source of information for tree detection. This paper presents a new method for tree detection based on high points’ densities from a high resolution Airborne LiDAR. The advantage of this method is that individual trees are detected based on the densities of high points which distinctively separates crown centers from crown edges. Therefore, regardless of the crown shape, the center of a crown has a higher density than the edge of the crown. The densities of high points for each point in a dataset are calculated in a column with a specified window size. At the beginning, all points in the dataset are selected as candidate point for tree locations. The tree locations are further refined by using Inverse Watershed segmentation in which higher weights will have better chances to be selected as tree locations than points with lower weight. The method is tested on different tree species and tree conditions for a floodplain area in the Netherlands. The results of the tree detection are compared with the actual tree locations. It is found that this method can correctly predict more than 70 percent of trees under different tree conditions. This method is sensitive to the density of undergrowth vegetation, vegetation type, size of trees, and density of crown cover caused by overlapping tree crowns. Further work is required on using this information to optimize this method.