Silvilaser 2019 - Poster Presentations »
Multiscale 3D-windows for detecting dead standing Eucalyptus from voxelised full-waveform LiDAR data
Dead Eucalyptus camaldulensis are an important for preserving biodiversity in native Australian forests. Nevertheless, detecting Eucalypt trees from LiDAR data acquired from native Australian forest is a challenge due to its big spatial resolution, the variance density of trees and the high standard deviation between tree heights. Most studies first perfom tree crown delineation and afterwards classify trees as dead or alive. Tree crown delineation is usually performed by detecting local maxima from the canopy height model (CHM) and then segmenting trees using the watershed algorithm, but Eucalypt trees has multiple trunk splits making tree delineation difficult. Shendryk et al, 2016, published an interesting Eucalyptus delineation algorithm that performs segmentation from bottom to top, but pulse density was 36 points/m2 around forested areas (expensive to acquire for big spatial resolution). This study improves the approach presented at Miltiadou et al, 2018, where it was showed that detection of dead standing Eucalypt trees without tree delineation is possible. Miltiadou et al, 2018, uses 3D windows to extract composite information characterising dead trees and then performs machine learning approaches (weighted-distance KNN with Random Forest) for 3D object detection (dead trees). This study introduces detection of dead standing Eucalypt trees using multi-scale 3D-windows for extracting composite information and consequently tackling height variations. By cross validating the algorithm impemented, it is shown that the multi-scale 3D-windows approach improved the precision (TP/(TP+FP)) and recall (TP/(TP+FN)) of predication.