Silvilaser 2019 - Poster Presentations »
COMPARING THE NUMBER OF TREES RETRIEVED WITH BOTH ALS AND TLS POINT CLOUD OVER A LOBLOLLY PINE STAND
The aim of this work was to isolate 16 years old loblolly pine trees, in a 400-m² circular plot, by means of two different methods of data collection: Airborne Laser Scanner (ALS) and Terrestrial Laser Scanner (TLS) data and compare them/. The ALS data were collected at flight height of 853 m, with a scan angle of 11° and 65% of overlap. The density of point cloud is approximately 22 points.m-2. The TLS data were collect with a Leica ScanStation P40 equipment with a scan resolution of 1 point at each 6.3 mm. We performed 5 scan stations at plot level, one with 360° in the center of the plot and 4 other on the corners with 110°. This configuration was adopted to avoid shading between the plot trees and ensure multiple scan data. Reference targets were distributed at the scan moment to ensure the geometric registration of each single scan in a multiple scan data. The ALS data processing consisted of filtering the point cloud by using Adaptive Triangular Irregular Network (ATIN) filter to separate the terrain and vegetation points. Next, Digital Terrain Model (DTM) and Digital Surface Model (DSM) were obtained by interpolating the filtered points, and normalizing them to obtain the Canopy Height Model (CHM). The CHM was smoothed with different search windows to emphasize the treetops pixels, and with this, the Thiessen’s Polygons segmentation was used to isolate the trees crowns. As TLS data processing, the steps were: (i) point cloud registration and delimitation, (ii) point cloud cut in a cross section with a certain height to obtain only the points relative to the tree stems, (iii) application of Label Connected Components algorithm to eliminate the points that not represent the tree stems, and (iv) segmenting them from octree and minimum number of points parameter. With the segmented tree, the tree detection was performed by assuming a threshold in the segments and the plantation alignment which resulted in vectors that were classified as stem, non-stem and uncertain. The evaluation of the applied methods was performed based on the counting of the trees in the field and calculated the commission and omission errors and accuracy. With the ALS data, the accuracy of detected trees was 60.38%, with an omission error of 30.19% and commission error of 9.43%. In TLS data, the automatic method for tree detection had an accuracy of 96.23% with an on omission error of 3.77%. The lack of paring and silvicultural treatment increased the amount of suppressed trees and many tree brunches, making it harder to identify the treetops in ALS data as well as the trees shading, interfering the stem identification in TLS point cloud. These factors affected negatively the automatic identification mainly in ALS points cloud. This study was sponsored by CAPES, CNPq, FAPESC and MANFRA. The ALS data were provided by the Sustainable Landscape Project.