Abstract
Virtual laser scanning (VLS), the simulation of laser scanning in a computer environment, is as a useful tool for field campaign planning, acquisition optimisation, and development and sensitivity analyses of algorithms in various disciplines including forestry research. One key to meaningful VLS is a suitable 3D representation of the object of interest. For VLS of forests, the way trees are constructed influences both the performance and the realism of the simulations. In this contribution, we analyse how well VLS can reproduce scans of individual trees in a forest. Specifically, we examine how different voxel sizes used to create the virtual forest affect point cloud metrics (e.g. height percentiles) and tree metrics (e.g. tree height and crown base height) derived from simulated point clouds. The level of detail in the voxelisation is dependent on the voxel size, which usually influences the number of voxel cells of the model. A smaller voxel size (i.e., more voxels) increases the computational cost of laser scanning simulations but allows for more detail. We present a method that decouples voxel grid resolution from final voxel cube size by scaling voxels to smaller cubes, whose surface is proportional to estimated normalised local plant area density. Voxel models are created from terrestrial laser scanning point clouds and then virtually scanned in one airborne and one UAV-borne simulation scenario. Using a comprehensive dataset of spatially overlapping terrestrial, UAV-borne and airborne laser scanning field data, we compare metrics derived from simulated point clouds and from real reference point clouds. Compared to voxel cubes of fixed size with the same base grid size, using scaled voxels greatly improves the agreement of simulated and real point cloud and tree metrics. This can be largely attributed to reduced artificial occlusion effects when using scaled voxels. The scaled voxels better represent gaps in the canopy, allowing for higher and more realistic crown penetration. Similarly high metric accuracy can be achieved using regular fixed size voxel models with notably finer resolution, e.g. 0.02 m. But this can pose a computational limitation for running simulations over large forest plots due to the ca. 50 times higher number of filled voxels. We conclude that opaque scaled voxel models enable realistic laser scanning simulations in forests and avoid the high computational cost of small fixed size voxels. -- This dataset includes - HELIOS++ data files to reproduce the simulations (forest (voxel) models, scene files and survey files) - Reference point clouds of the forest plots and of individual target trees - Simulated point clouds of the forest plots and of individual target trees - TLS point clouds of the target trees, based on which the trees were reconstructed for the simulations - Metrics computed from the reference and the simulated tree point clouds - Python scripts used for voxel processing, simulation output processing and metric computation - Examples of configuration files for AMAPVox (https://amap-dev.cirad.fr/projects/amapvox), the software for voxel-based plant area density estimation
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have