Orientation of Image Sequences Acquired from UAVs and with GPS Cameras

摘要

This paper presents an approach for automatically orienting images taken with different cameras from unknown positions. Specifically, we show that we can relatively orient wide-baseline images taken from the ground as well as from lightweight (micro) unmanned aerial vehicles (UAVs) of about 1 kg overall weight. The ground images were collected using a consumer GPS camera which allows us to upgrade relatively oriented 3D models to absolute orientation. Automatic relative orientation based on points as well as absolute orientation with consumer GPS cameras both suffer from possibly extreme outliers. We tackle this challenge by combining consensus-based filtering with robust adjustment. Results obtained with the developed procedure demonstrate the potential of our approach.