Comparison of Dense Stereo Using CUDA

In this paper, a local and a global dense stereo matching method, implemented using Compute Unified Device Architecture (CUDA), are presented, analyzed and compared. The purposed work shows the general strategy of the parallelization of matching methods on GPUs and the tradeoff between accuracy and run-time on current GPU hardware. Two representative and widely-used methods, the Sum of Absolute Differences (SAD) method and the Semi-Global Matching (SGM) method, are used and their results are compared using the Middlebury test sets.

[1]  Allen R. Hanson,et al.  Computer Vision Systems , 1978 .

[2]  Heiko Hirschmüller,et al.  Evaluation of Stereo Matching Costs on Images with Radiometric Differences , 2009, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[3]  Peter Reinartz,et al.  Analysis of Image Sequences for the Detection and Monitoring of Moving Traffic , 2009 .

[4]  Stefan K. Gehrig,et al.  A Real-Time Low-Power Stereo Vision Engine Using Semi-Global Matching , 2009, ICVS.

[5]  Mohammed Bennamoun,et al.  Non-parametric techniques for fast and robust stereo matching , 1997, TENCON '97 Brisbane - Australia. Proceedings of IEEE TENCON '97. IEEE Region 10 Annual Conference. Speech and Image Technologies for Computing and Telecommunications (Cat. No.97CH36162).

[6]  Heiko Hirschmüller,et al.  Stereo Processing by Semiglobal Matching and Mutual Information , 2008, IEEE Trans. Pattern Anal. Mach. Intell..

[7]  A. Verri,et al.  A compact algorithm for rectification of stereo pairs , 2000 .

[8]  Richard Szeliski,et al.  A Taxonomy and Evaluation of Dense Two-Frame Stereo Correspondence Algorithms , 2001, International Journal of Computer Vision.