文章基本信息

标题：Inventory Using Lidar and Aerial Images for 3D Treetop Positioning, Species Recognition, Height and Crown Width Estimation
本地全文：下载
作者：I. Korpela ; B. Dahlin ; H. Schäfer 等
期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2007
卷号：XXXVI-3/W52
页码：227-233
出版社：Copernicus Publications
摘要：An entire single-tree remote sensing (STRS) system was developed and tested in an inventory of timber resources of a 56.8-hectare forest. A semi-automatic approach with operator intervention is used in the system and it solves the essential tasks of STRS: 3D treetop positioning, height estimation, species recognition, crown width estimation and the model-based allometric estimation of the stem diameter. Large-scale aerial imagery, an accurate DTM and semi-dense LiDAR data are required. The relatively low sampling density of the LiDAR, 6 points per m 2 here, was considered appropriate for crown width estimation, when the 3D treetop position, tree height estimation and species classification are done first using the images. LiDAR-based crown width estimation was done using crown modeling, in which parametric crown instances are iteratively fitted with the LiDAR data. Image-based 3D treetop positioning and crown width estimation rely on multi-scale template matching (MSTM). Species recognition was done by visual photo-interpretation. In the experiment, a total of 59 circular 0.04-ha plots and 5294 trees were measured using STRS. The plots were investigated in the field and all STRS-trees and omission trees with a stem diameter of above 50 mm were mapped. The mapping was based on the use of the STRS-trees as geodetic control points. Redundant intertree azimuth and distance observations and a weighted least square adjustment of observations was used for the positioning of the omission trees. The commission error-rate was 2% in stem number and the omission trees constituted 10% of the total stem volume. Visual species recognition accuracy was 95% in classes of pine, spruce, broadleaved and dead trees. Height estimation accuracy of MSTM was 0.71 m or 4.7% in RMSE and it includes the DTM-errors. Stem diameter estimation RMSE was 29% and 20% when the crown widths were estimated using images and LiDAR, respectively. Underestimation of stem diameters was considerable, 3.4 and 1.0 cm. The inaccuracy of the stem diameter estimates degraded the accuracy of single-tree volume estimates and the results of estimating the proportion of assortments. Calibration of the STRS measurements and estimates are needed and this calls for field observations
关键词：Allometry; Modeling; Mapping; Sampling; Photo-plot; Calibration; Multi-Scale; Matching; Template