Single tree detection in forest areas with high-density LIDAR data

The study presents a novel method for delineation of tree crowns and detection of stem positions of single trees from dense airborne LIDAR data. The core module of the method is a surface reconstruction that robustly interpolates the canopy height model (CHM) from the LIDAR data. Tree segments are found by applying the watershed algorithm to the CHM. Possible stem positions of the tallest trees in the segments are subsequently derived from the local maxima of the CHM. Additional stem positions in the segments are found in a 3-step algorithm. First, all the points between the ground and the crown base height are separated. Second, possible stem points are found by hierarchical clustering these points using their horizontal distances. Third, the stem position is estimated with a robust RANSAC-based adjustment of the stem points. We applied the method to small-footprint full waveform data that have been acquired in the Bavarian Forest National Park with a point density of approximately 25 points per m². The results indicate that the detection rate for coniferous trees is 61 % and for deciduous trees 44 %, respectively. 7 % of the detected trees are false positives. The mean positioning error is 0.92 cm, whereas the additional stem detection improves the position on average by 22 cm.