Assessment of Tree Attributes Extraction Algorithms

Abstract

Canopy Height Model (CHM)-based and point-based tree extraction algorithms are two common techniques to extract individual trees from airborne lidar data. In general, point-based algorithms process lidar points directly but suffer from intensive computation while CHM-based algorithms are efficient but fail to extract sub-dominant trees in dense canopies and their performances rely on the resolution of the CHM. To test which existing algorithm yields most accurate results, this paper compares Fixed-window local maxima algorithm, Popescu and Wynne's local maxima algorithm, variable area local maxima algorithm, individual-tree-crown delineation algorithm (ITC) and Li's point-based segmentation algorithm (LPS). The comparisons of the results to the reference data indicate although LPS extracts the largest number of individual trees, the extracted tree heights and crown widths are less accurate since LPS suffers from oversegmentation. In contrast, ITC outperforms other algorithms when measured by extracted tree heights and crown widths due to the adaptive window size. To testify if CHMs with better resolution promote the accuracy of extracted tree heights and crown widths since they provide more details, CHM-based algorithms are also applied to CHMs with different resolutions. The results indicate that although CHMs with better resolution help to yield more accurate tree heights, they do not necessarily result in more accurate crown widths.