Orientation of Ground-level Motion Imagery Using Building Facades

In this paper, we address the orientation of ground-level motion imagery captured by sensors roaming in an urban area. We investigate the use of building facades (instead of traditional points), as matching features for ground-level motion imagery orientation. We assume a situation where few images in our sequence are already absolutely oriented, and present a novel approach to orient all remaining in-between image sequences relative to them. This innovative version of dependent orientation allows us to propagate orientation information within sequences of ground level imagery, establishing a novel orientation scheme. Experimental results show accuracies on the order of 0.18 to 0.29 degrees in rotation estimation, and 0.17 to 0.29 meters in camera position determination.