Robust ground peak extraction with range error estimation using full-waveform LiDAR
Abstract
Topographic mapping is one of the main applications
of airborne LiDAR. Waveform digitization and processing
allow for both an improved accuracy and a higher ground
detection rate compared to discrete return systems. Nevertheless,
the quality of the ground peak estimation, based on last
return extraction, strongly depends on the algorithm used. Bestperforming
methods are too computationally intensive to be used
on large datasets. We used Bayesian inference to develop a new
ground extraction method whose most original feature is predictive
uncertainty computation. It is also fast, and robust to ringing
and peak overlaps. Obtaining consistent ranging uncertainties is
essential for determining the spatial distribution of error on the
final product, point cloud or DEM. The robustness is achieved by
a partial deconvolution followed by a Bayesian Gaussian function
regression on optimally truncated data, which helps reduce the
impact of overlapping peaks from low vegetation. Results from
real data are presented, and the gain with respect to classical
Gaussian peak fitting is assessed and illustrated.