Camera motion tracking in a dynamic scene

To insert a virtual object into a real image, the position of the object must appear seamlessly as the camera moves. This requires camera tracking with estimations of all internal and external parameters in each frame with an adequate degree of stability to ensure negligible visible drift between the real and virtual elements. In the post production of film, matchmoving software based on SfM is typically used in the camera tracking process. However, most of this type of software fails when attempting to track the camera in a dynamic scene in which a moving foreground object such as a real actor occupies a large part of the background. Therefore, this study proposes a camera tracking system that uses an auxiliary camera to estimate the motion of the main shooting camera and 3D position of background features in a dynamic scene. A novel reconstruction and connection method was developed for feature tracks that are occluded by a foreground object. Experimentation with a 2K sequence demonstrated the feasibility of the proposed method.