Registration of UAV images

From TEILab
Jump to: navigation, search

Project Team

  • Dr. Francis Quek Professor, Center for HCI
  • Yingen Xiong Graduated-PhD student
  • Chen Li Graduated-PhD student
  • Travis Rose Graduated-PhD student
  • R. K. Hill & Andy Drake, MTL, Systems Inc.

Note: This is a past project that is currently dormant

Project Overview


In this project, we investigate the registration of multi-position images taken from unmanned aerial vehicles (UAVs). Depending on the elevation and azimuth of the aircraft and camera system, these images may exhibit a variety of perspective distortions (typically keystoning), and the orientation, scale and position of the images are not available. Hence, to register such images, one has to determine the affine transformation between the images (scale, position and rotation).

In our current 'eigen disk' approach, we extract points of high spatial content from the disparate images and extract epsilon disks around them. The eigen vectors of these disks are invariant to orientation and translation. Hence, we can employ a multi-scale voting scheme in eigen space to determine the relative orientations of sub-images from different UAV images. We have demonstrated the accuracy of this registration scheme. Since the keystone distortions between any two images are conformal, we are investigating approaches to interpolate and warp the two images so that they can be registered.


This project was supported by the United States Air Force STTR Program.