Share Email Print
cover

Proceedings Paper

Optimization of semi-global stereo matching for hardware module implementation
Author(s): Mikołaj Roszkowski
Format Member Price Non-Member Price
PDF $14.40 $18.00

Paper Abstract

Stereo vision is one of the most intensively studied areas in the field of computer vision. It allows the creation of a 3D model of a scene given two images of the scene taken with optical cameras. Although the number of stereo algorithms keeps increasing, not many are suitable candidates for hardware implementations that could guarantee real-time processing in embedded systems. One of such algorithms is semi-global matching, which seems to balance well the quality of the disparity map and computational complexity. However, it still has quite high memory requirements, which can be a problem if the low-cost FPGAs are to be used. This is because they often suffer from a low external DRAM memory throughput. In this article, a few methods to reduce both the semi-global matching algorithm complexity and memory usage, and thus required bandwidth, are proposed. First of all, it is shown that a simple pyramid matching scheme can be used to efficiently reduce the number of disparities checked per pixel. Secondly, a method of dividing the image into independent blocks is proposed, which allows the reduction of the amount of memories required by the algorithm. Finally the exact requirements for the bandwidth and the size of the on-chip memories are given.

Paper Details

Date Published: 25 November 2014
PDF: 12 pages
Proc. SPIE 9290, Photonics Applications in Astronomy, Communications, Industry, and High-Energy Physics Experiments 2014, 92902T (25 November 2014); doi: 10.1117/12.2075012
Show Author Affiliations
Mikołaj Roszkowski, Warsaw Univ. of Technology (Poland)


Published in SPIE Proceedings Vol. 9290:
Photonics Applications in Astronomy, Communications, Industry, and High-Energy Physics Experiments 2014
Ryszard S. Romaniuk, Editor(s)

© SPIE. Terms of Use
Back to Top