Robust Estimation of Camera Parameters Pan, Tilt and Zoom for Integration of Virtual Objects into Video Sequences

The integration of virtual objects into an image sequence captured by a moving camera is of high interest for special eeects in tv-and lm-productions. For the composition, the camera parameters of the real camera are estimated and transferred to a virtual camera. The estimation accuracy is very critical, because even small estimation errors can be recognised as localisation jitter of the inserted virtual objects. This contribution presents a robust estimation technique for computation of camera parameters pan, tilt and zoom from an image sequence. Based on conventional feature detection and correspondence analysis, the robustness of the approach is achieved by using a highly eecient outlier detection of misaligned correspondences.