The aim of the project is to determine the trajectory of a drone in flight, by observing it with two or more cameras. Camera-based tracking provides a ground-based al-ternative to trajectory estimation from on-board GNSS+IMU measurements. To obtain a flexible setup we aim for a system that does not require pre-calibrated camera poses or synchronised image sequences.
The first task is to detect the drone in the image using pattern recognition tech-niques. When the image trajectories are detected, the epipolar geometry and time shift between pairs of cameras can be computed using existing methods that are available in the group. The pairwise camera poses and time offsets then need to be combined and optimization task formulated such that global spatio-temporal calibra-tion is found across all cameras. From the camera calibration, 3D trajectory of the drone can be triangulated. Afterwards, an optimization step should be employed to refine the trajectory.
The thesis will cover a wide variety of computer vision techniques, including object tracking and 3D geometry. The individual tasks, however, will build upon existing methods and should not be very complicated, making it an ideal opportunity to get hands-on experience in building a 3D computer vision system. If successful, the method will have considerable potential to be practically used in both research and industry.
Cenek Albl (firstname.lastname@example.org)
3D Computer Vision System
CLS Student Project (MPG ETH CLS)
Information, Computing and Communication Sciences
Engineering and Technology