Tandem Flight Quadcopters for Wide Baseline Stereo 3D Reconstruction

In this work we explore the use of two micro-unmanned aerial vehicles (MAVs) in tandem flight for 3D reconstruction. Having two distinct viewpoints simultaneously helps to deal with the 3D reconstruction of dynamic scenes, which is not possible with monocular structure from motion. Having the leader MAV in sight of the follower enables visual servoing to have a steady baseline. A ground station is used to run trackers on the image streams. A gyro-aided Luca Kanade tracker is used, running on a GPU using CUDA based code. These can then be combined to have a more precise structure from motion sparse depth map, and a more coarse stereo based dense depth map that enables segmenting independent motion in the sparse map. The two viewpoints can also be combined in a panorama image and the trackers run on this wider image, so as to keep features in sight longer.

Supervisor: Prof. Jorge Lobo