Stereo Vision System Implemented on FPGA

Abstract

Stereo vision systems are computer vision implementations that use stereoscopic ranging techniques to estimate a 3D model of a scene. Stereo vision employs triangulation, a popular ranging technique, to compute the depth from 2D images. The theoretical background and geometry that associates visual disparity to scene structure is obvious; however, the procedure to automatically compute disparity by inter-image correspondence in real-time is challenging. This paper proposes a stereo vision system that provides dense depth maps in real-time from monochrome cameras. The entire process is realized in a single field programmable gate array (FPGA). FPGAs, being inherently parallel in nature, are quite suitable for the problem and are found to be much faster for such computer vision applications from previous studies.