Silvilaser 2019 - Poster Presentations »
Improving ground filter optimisation using terrestrial point clouds
Accurate measurements of vegetation extent and structure are vital for a range of ecological assessments; ecological monitoring, precision forestry, carbon accounting and fuel and fire behaviour estimates. Active and passive sensors have been utilised to capture 3D point clouds to be subsequently analysed describing vegetation health, structure and biomass. Height is a common metric to extract from these 3D representations. As such, the extraction of height requires the use of a ground filter to determine the location of the underlying terrain. Prior research has highlighted that despite point clouds containing large amounts of information describing the ground, the extraction of this information and optimisation of filter settings to the required accuracy can be challenging particularly in forested areas with complex vertical structure and undulating terrain. This research presents a new approach for optimising the ground filtering process in a forested environment. A suite of point clouds were captured using both passive and active sensors over a 50 x 30m plot. Image based point clouds were captured from the ground using a Sony A6000 that was converted to capture reflectance in the Near Infrared (NIR)and a DJI Phantom 4 Pro for airborne capture. A Trimble TX-8 and Velodyne Puck were used to capture LiDAR data. The terrestrial NIR point clouds were captured as a 20m transect with a reflectance threshold applied to characterise the ground surface. This estimate of ground was validated against reference ground measurements in four 0.5mx0.5m quadrats placed along two transects. The NIR image based point cloud was subsequently used as the reference for optimising settings for the cloth simulation, adaptive TIN and progressive morphological filters for the TLS and airborne point clouds minimising for rmse. Spot heights were sampled across the plot for ground control. Results indicate that using spot height measurements for filter parameterisation often results in significantly different parameters to using the proposed transect approach. Whilst, UAV SfM and UAV LiDAR had sufficient information content in this environment to characterise a ground surface, the variations in filter parameterisation are sufficient to cause large differences in metrics describing understorey vegetation (height, volume and biomass). With terrestrial and airborne methods of 3D data capture being increasingly used to observe and monitor fine scale change in vegetated landscapes, the method presented may be used to determine a ground surface that allows a more accurately characterisation of these forest strata.