New Parallel Technics for GPU, Fast SURF Algorithm
Hicham Hassnaoui1, Aisha Sahel2, Abdelmajid Badri3
1Hicham Hassnaoui, Department of Computer vision, Faculty of Sciences and Techniques, Mohammedia, University Hassan Casablanca, Morocco.
2Aicha Sahel, Professor, Department of Computer vision, Faculty of Sciences and Techniques, Mohammedia, University Hassan Casablanca, Morocco.
3Abdelmajid Badri, Professor, Department of Electrical Engineering, University of Hassan of Casablanca.
Manuscript received on May 16, 2020. | Revised Manuscript received on May 26, 2020. | Manuscript published on June 10, 2020. | PP: 389-393 | Volume-9 Issue-8, June 2020. | Retrieval Number: 100.1/ijitee.G5855059720 | DOI: 10.35940/ijitee.G5855.069820
Open Access | Ethics and Policies | Cite | Mendeley
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: Computer vision algorithms, especially real-time tasks, require intensive computation and reduced time. That’s why many algorithms are developed for interest point detection and description. For instance, SURF (Speeded Up Robust Feature) is extensively adopted in tracking or detecting forms and objects. SURF algorithm remains complex and massive in term of computation. So, it’s a challenge for real time usage on CPU. In this paper we propose a fast SURF parallel computation algorithm designed for Graphics-Processing-Unit (GPU). We describe different states of the algorithm in detail, using several optimizations. Our method can improve significantly the original application by reducing the computation time. Thus, it presents a good performance for real-time processing.
Keywords: Computer vision, GPU, Parallel computation, SURF, Tracking.
Scope of the Article: VLSI Algorithms