Flexible Architecture for Three Classes of Optical Flow Extraction Algorithms

Abstract

Recent findings have shown that several visual perceptual tasks such as 3D navigation and visual attention could use optical flow as a key step in their complex computing models and mechanisms for visual perception of motion and dynamic scene analysis. We have been performing experimental studies to gain understanding of the general principles in the optical flow computation under different approaches so as to establish efficient implementation techniques of embedded systems able to perform high speed motion estimation. Based on these studies, a custom parallel hardware architecture for three classes of optical flow estimation algorithms has been proposed and mapped into a digital implementation. A flexible field programmable gate array (FPGA) implementation is presented witch allow to compute the optical flow at a rate of 50 frames per second with 200times200 gray level images with low resource utilization.