Active sensors for UAV autonomous navigation on Amazon region

Abstract

This work is an additional exploration inspired by the results of an earlier study of the geo-localization problem over a densely forested region of the Brazilian Amazon forest. Light Detection and Ranging (LiDAR) data was post-processed from 3D cloud point format to 2D elevation images and template matching was used with normalized cross-correlation. Within a constrained search area it was possible to geo-localize the 2D patches of surface images on Interferometric Synthetic Aperture Radar (InSAR) elevation data. The transect 3D cloud point was transformed into a 12.5m resolution 2D surface image with the circular binning procedure, a resolution compatible with the Advanced Land Observation Satellite (ALOS) elevation maps used as reference. This application of template matching achieved 36m root mean square error, or about 4 pixels of error, over the LiDAR transect route. Position estimation is essential for autonomous navigation of aerial vehicles, and experiments with LiDAR data show potential for localization over densely forested regions, where Computer Vision methods using optical camera data may fail to acquire distinguishable features.