Automatic Tree Annotation in LiDAR Data

Monitoring the dune-beach system of Guardamar del Segura (Spain) using UAV, SfM and GIS techniques

“…The presence of vegetation on foredunes affects the photogrammetric reconstruction of the terrain, and if not masked its seasonal changes may generate a spatial difference that can be considered as a volume change in the measurements obtained by the difference surface tool (Scarelli et al, 2017). In this work the vegetation cover has been manually masked, but there are algorithms and filtering techniques to automatically identify the vegetation and create DSM instead of DEM (Gupta et al, 2018).…”

Tree Annotations in LiDAR Data Using Point Densities and Convolutional Neural Networks

“…The first method, MultiReturn [23], which can be reformulated in Algorithm 1, is based on four distinct steps:…”

“…The first method, termed as MultiReturn, is based on our earlier work [23] and uses traditional handcrafted features as well as inherent data characteristics of LiDAR data. It works well on datasets that have point cloud density >20 points m −2 as these datasets contain the number of returns characteristic that allows the method to identify trees.…”

“…We adapt the training loss to deal with unbalanced class distributions, allowing the network to train a binary classifier with very few positive points (tree category) relative to large negative points (non-tree points). The first method, MultiReturn [23], which can be reformulated in Algorithm 1, is based on four distinct steps:…”

“…Map of survey area with tree locations shown in yellow and the outputs of our MultiReturn[23] labelling algorithm shown in red.…”