| A technique to increase the frame rate of digital video cameras at high-resolution is presented. The method relies on special video hardware capable of simultaneously generating low-speed, high-resolution frames and high-speed, low-resolution frames. The algorithm follows an estimate-and-correct approach, in which a high-resolution estimate is first produced by translating the pixels of the high-resolution frames produced by the camera with respect to the motion dynamic observed in the low-resolution ones. The estimate is then compared against the current low-resolution frame and corrected locally as necessary for consistency with the latter. This is done by replacing the wrong pixels of the estimate with pixels from a bilinear interpolation of the current low-resolution frame. Because of their longer exposure time, high-resolution frames are more prone to motion blur than low-resolution frames, so a motion blur reduction step is also applied. Simulations demonstrate the ability of our technique in synthesizing high-quality, high-resolution frames at modest computational expense. |
