Against high computational cost and severe accumulative error problems existing in the current video mosaicing methods, we propose a real-time mosaicing algorithm for aerial videos based on geographical parameters. First, we decompose the camera motion model for image registration in real time. We then estimate the principal orientation of each frame, and extend the uni-orientation manifold mosaicing algorithm to a multi-orientation algorithm. Finally, with geographical parameters, we rectify mosaicing strips locally to make sure that the mosaics are mapped precisely on the real geographical positions. Experimental results for synthetic and real videos demonstrate accuracy and the near real-time property of our algorithm, with the mosaicing speed up to 25 frames per second for videos of 640£480 resolution.
