Abstract
Full-pixel motion estimation is conducted in real time at 500 fps using GPU acceleration of a gradient-based optical flow algorithm. Independently of cell-size in calculating product sums of brightness gradients, the improved Lucas-Kanade method can be remarkably accelerated for fast motion estimation by calculating integral images of brightness gradients in parallel. The improved algorithm is implemented on a GPU-based high-speed vision platform, and optical flow estimation can be executed in real time at 500 fps for 512×512 images. Its performance was verified for real scenarios such as rapid human motion.
