Image mosaicing with quadricopter AR.Drone
Pedro damian Kostelec · 2013
In this bachelor thesis we developed a system for the automatic mosaicing of video frames obtained from the mobile platform Parrot AR.Drone. Sequential images were collated by extracting putative matching keypoints. Matching pairs were filtered with the RANSAC algorithm and the estimated homography was optimized using the Levenberg-Marquardt non-linear optimization method. We also present a set of rules to detect badly estimated homographies. We evaluate the generated mosaics and explain why some images are better suited for mosaicing than others. Moreover, we attempted to reconstruct the flight path using navigation data received from the quadricopter during flight. We concluded that the measured navigation data is too imprecise and noisy to use it for mosaicing.