BayesMap at AGU 2014 in San Francisco
We will be at at the AGU Fall Meeting, San Francisco, 15-19 Dec 2014

Abstract Title: Automated Probabilistic LiDAR Swath Registration

Final Poster Number: EP41B-3539

Session Title: EP41B. High-Resolution Topography and Process Measurements for Analyzing Earth-Surface Dynamics I Posters

Day/Time: Thursday, 18 December 2014: 08:00 AM – 12:20 PM, MS, Poster Hall


We recently developed a new point cloud registration algorithm. Compared to Iterated Closest Point (ICP) techniques, it is robust to noise and outliers, and easier to use, as it is less sensitive to initial conditions. It minimizes the entropy of the joint point cloud (including intensity attributes to help register areas with poor relief), uses a voxel space and B-Spline interpolation to accelerate computation.

A natural application of registration is swath alignment in airborne light detection and ranging (LiDAR). Indeed, due to uncertainty in the inertial navigation system (INS), attitude angles are subject to time-dependent errors. Such errors can be understood as a sum of three terms: a global term, or boresight error, which can be addressed using several existing techniques; a low-frequency term, which is modeled as a constant attitude error for regions several hundred meters along-track; a high-frequency term, responsible for corduroy artifacts, not addressed here. We propose to use the new registration algorithm to correct the low-frequency attitude variations. Relative geometric errors are significantly reduced, as pairs of swaths are registered onto each other local corrections. Absolute geometric errors are reduced during a second step, by applying all the corrections together to the entire dataset.

We used a test area of 200 km2 in Portugal, with a density of 3-4 pts/m2. The point clouds were derived from waveform data, and include predictive range uncertainties estimated within a Bayesian framework. The data collection was supported by FCT and FEDER as part of the AutoProbaDTM research project (2009-2012).

Modeling and reducing geometric error helps build consistent uncertainty maps. After correction, residual errors are taken into account in the final 3D error budget. For gridded elevation models a vertical uncertainty map is computed.

Finally, it is possible to use the inter-swath registration parameters to estimate the distribution of horizontal error, which is usually difficult to assess.