Camera parameter estimation from a long image sequence by tracking markers and natural features

Abstract

Abstract Camera parameter recovery from an image sequence is very important in many applications such as 3D model reconstruction, object recognition, robot navigation, and mixed reality. However, there is a problem concerning the precision in camera parameter estimation from the image sequence, because feature tracking errors and estimation errors are accumulated. In this paper, we propose a camera parameter estimation method which is based on using a number of markers with known 3D position, color, and shape, as well as natural features. Initially, the camera parameters and 3D positions of natural features are estimated efficiently in every frame by tracking both these markers and natural features. The accumulation of estimated errors is then minimized by specifying additional markers in some frames of input and optimizing parameters globally through the whole input. In the experiments, some results of 3D reconstruction show the validity of the proposed method. © 2004 Wiley Periodicals, Inc. Syst Comp Jpn, 35(8): 12–20, 2004; Published online in Wiley InterScience ( www.interscience.wiley.com ). DOI 10.1002/scj.10702