View on GitHub

Dynamic Baseline Binocular Stereo Using Multirotor UAVs

Undergrad thesis 2015 (Electrical Engineering) by Ankush Gola and Joseph Bolling. A dynamic baseline binocular stereo vision system in which the perspectives are shot from independent UAVs. Advised by Professors Andrew Houck and Peter Ramadge.

Download this project as a .zip file Download this project as a tar.gz file

Abstract

We design and implement a 3D film-making system in which the left and right perspectives are shot from independent unmanned aerial vehicles (UAVs), thereby producing a stereo vision system with a dynamically variable baseline and elevated perspective. This system allows for novel visual effects due to the amplified depth perception achieved from the wide baseline. Furthermore “shrinking” and “growing” effects are achievable by dynamically varying the baseline. Our designs are motivated by research into human stereopsis and existing work in computer vision. We discuss our UAV platform design, techniques for UAV control and synchronization, current progress in enhancing vehicle state estimation through vision-based UAV localization, and our video stabilization pipeline. We achieve video that is comfortably viewable and lay the groundwork for further system enhancements.

FULL PAPER

Videos

Formation Flight Test

Autonomous UAV flight test showcasing "moving apart" formation. Landing is manually controlled.

3D Footage (No Stabilization)

Raw stereo footage in anagplyh form with NO video post-processing.

3D Footage

Stereo footage in anagplyh form WITH video post-processing (sensor fusion, state estimation, camera frame corrections).