Marker-less Real-time Camera Registration for Mixed Reality

Abstract

A real-time and robust algorithm for 3D camera registration in a Mixed Reality (MR) environment is described in this paper. The most used technique for camera pose (position and orientation with respect to a fixed or moving object) is based on fiducial marker tracking. This method guarantees good results in real-time with a single camera, but needs several high contrast printed markers on external world in order to make possible the calculation of camera parameters and positioning. Thus real 3D geometric data are grabbed only through already known markers. The aim of this research is a real-time monocular camera tracking and registration through automatic image features extraction from video streaming. The first implementation of the method, several examples and confrontation with non interactive algorithm for SFM (Structure From Motion) have demonstrated that this meets the real-time response and sufficient precision needed by a Mixed Reality environment.