Real-time Stereo for Embedded Vision System

Abstract

This paper describes a real-time stereo vision system for mobile embedded vision applications. An NCC based cost function with a rectangular mask is used to evaluate corresponding pixels, which minimize matching cost energy. We propose a hardware-friendly architecture to reducing the computation complexity in a matching algorithm and implement it on a field programmable gate array (FPGA). The proposed data reuse technique makes it possible to be 20 times faster than a simple parallel processing method and it also occupies less hardware resources than previous works. The proposed stereo vision system is expected to be an efficient solution for a low cost and real-time embedded vision system.