공통의 시야를 갖지 않는 다수의 RGB-D 카메라간의 외부 파라미터 추정방법

Abstract

In order to estimate extrinsic parameters for multiple cameras, we use chessboard patterns. However, it is not available when multiple cameras do not have common field of views simultaneously. In this paper, we propose an extrinsic camera calibration method using visual features for multiple cameras which do not have common field of views. We adopt SLAM techniques to build a map from individual cameras. Then we apply bundle adjustment to align the maps resulting robust estimation of extrinsic parameters of the cameras. We provide experimental results showing improved performance in terms of reprojection error.