Fast and Accurate Optical Flow Estimation using FPGA

Abstract

In this paper, we extend an approach used in the stereo vision for the optical flow estimation to achieve lower error rates. In the optical flow estimation, two dimensional search is required, and more hardware resources becomes necessary than the stereo vision that requires only one dimensional search. In our implementation, the target image is divided into sub-images, and they are processed in turn to reduce the required circuit size. The error rates by our system is much lower than previous works, and its processing speed is fast enough for practical applications.