Tree Top Detection in UAV Data: Evaluating Accuracy of Different Estimation Techniques

Abstract

The tree top point position is important for calculating many parameters and supporting various geometry and analyses. This study compares four methods, i.e., Local Maxima (LM), Template Matching (TM), Top Point Without Slope (TPWS) correction, and Top Point with Slope (TPS) correction, to estimate the tree top point position of Pinus eldarica using UAV-acquired RGB imagery (2 cm ground sampling distance) and high-density point clouds (1.27 points/cm3). The LM, and the TM methods are applied for estimating tree top point positions. The TPWS method uses the tree's shadow on terrain without slope correction, and finally, the fourth method uses the tree's shadow on the terrain with slope correction. Results were compared against Field Tree Top (FTT) point measurements. Findings reveal that LM and TPS were the most effective. LM provided the most accurate results overall, with a relative root-mean-square error (RRMSE) of 1.08, a mean error (ME) of 0.97, and a bias score (BS) of 0.23. Estimating the tree top point position with LM showed strong correlations (R2 = 0.94) with FTT position. This study underscores the value of LM and TPS methods for precise tree top point position estimation, highlighting the need for future research into the estimation of tree top point position methods.