Item

Abstract

This paper addresses the estimation of camera motion and 3D reconstruction from image sequences for multiple independently moving cameras. If multiple moving cameras record the same scene, a camera is often visible in another camera's field of view. This poses a constraint on the position of the observed camera, which can be included into the conjoined optimization process. The paper contains the following contributions: Firstly, a fully automatic detection and tracking algorithm for the position of a moving camera in the image sequence of another moving camera is presented. Secondly, a sparse bundle adjustment algorithm is introduced, which includes this additional constraint on the position of the tracked camera. Since the additional constraints minimize the geometric error at the boundary of the reconstructed volume, the total reconstruction accuracy can be improved significantly. Experiments with synthetic and challenging real world scenes show the improved performance of our fully automatic method.