Title :
A 3D local descriptor SHOT on massively parallel processors
Author :
Linjia Hu ; Nooshabadi, Saeid
Author_Institution :
Dept. of Comput. Sci., Michigan Technol. Univ., Houghton, MI, USA
Abstract :
This paper investigates the suitability of graphical process unit (GPU) for a novel 3D object local descriptor named Signature of histogram of orientations (SHOT). SHOT presents a great balance between descriptiveness and robustness but its high complexity incurs heavy computational workload. We designed two parallel SHOTs on GPU to speed up the original serial counterpart. Experiment results show both the precise and approximate parallel algorithms exhibit outstanding speed performance. Moreover, we validated the descriptiveness of both parallel SHOTs were at high level through quantitative experimental comparisons.
Keywords :
computer graphics; graphics processing units; parallel algorithms; 3D object local descriptor; GPU; SHOT; graphical process unit; massively parallel processors; parallel algorithms; signature of histogram of orientations; Acceleration; Algorithm design and analysis; Approximation algorithms; Graphics processing units; Histograms; Instruction sets; Three-dimensional displays;
Conference_Titel :
Consumer Electronics (ICCE), 2015 IEEE International Conference on
Conference_Location :
Las Vegas, NV
Print_ISBN :
978-1-4799-7542-6
DOI :
10.1109/ICCE.2015.7066373