January 2014
·
36 Reads
Terrain mapping using an unmanned aircraft is valuable in many different applications, such as those involving urban infrastructure classification, search and rescue, and post-disaster mapping where existing maps are now meaningless. LiDAR scanners provide one way of generating 3-D maps, however these devices are generally heavier, more expensive, and more power consuming than vision-based systems. Vision-based systems also offer the advantage of having color and texture information available, which can make substantial improvements to the classification accuracy. We propose using vision systems on small, unmanned aircraft to create 3-D terrain maps via Structure from motion and multi-view stereo methods. These maps are then used to classify the environment to aid in situational awareness, path planning or other actions that would benefit from intelligent object classification. We show that by training on a LiDAR dataset of an urban environment, we can achieve high accuracy when applied to our own smaller dataset generated using image-based 3-D reconstructions. © 2015 American Institute of Aeronautics and Astronautics Inc. All rights reserved.