Individual urban trees detection based on point clouds derived from UAV-RGB imagery and local maxima algorithm, a case study of Fateh Garden, Iran

As a remote sensing technique, unmanned aerial vehicles (UAVs) have great potential in several fields, such as monitoring vegetation in an urban area at low altitudes at a reasonable cost. In this study,we assessed the potential of individual tree detection by the structure of the motion algorithm (SfM) based on UAV images and derived point cloud. Urban broadleaved forests (Fateh garden) were photographed in the spring of 2018 with different structures, a mixed uneven-aged dense stand (MUDS), a mixed uneven-aged sparse stand (MUSS), and a pure even-aged dense stand (PDEs). The results of using the local maxima algorithm for the different structures showed a detection accuracy rate of 0.90, 0.54, and 0.32 for PDES, MUDS, and MUSS, respectively. Based on the results, the accuracy of tree detection is affected by the height of the trees (individuals with a height of fewer than 5 meters were not detected), and the species (Poplar trees were detected better than other species), as well as the searching window size. The fixed tree window size of 3x3 was the best window size, and the fixed smoothing window size was variable for each site. Using mean and Gaussian filters did not noticeably affect the results. In general, our study showed that the canopy height model (CHM) from UAV can detect trees with very high accuracy in urban forests with homogenous even-aged structures, while in uneven-aged stands, the accuracy of tree detection is medium to low.