Title :
Performance Evaluation of Feature Extraction Algorithm on GPGPU
Author :
Sawant, Namdev ; Kulkarni, Dinesh
Author_Institution :
Dept. of Comput. Sci. & Eng., Walchand Coll. of Eng., Sangli, India
Abstract :
Nvidia´s GPGPU based Compute Unified Device Architecture (CUDA) is a software platform for massively parallel high-performance computing on GPU. It provide several key abstractions- a hierarchy of thread block, shared memory, and barrier synchronization. This model has proven quite successful at programming multithreaded many core GPUs and scale transparently to hundreds of cores: many industry and academia are already using CUDA to achieve speedups on production and research codes. This paper analyze distinct feature of CUDA GPU, summarizes general programming mode of CUDA. This paper presents image processing algorithm i.e. Canny Edge detector which is used as pre-processing steps in many computer vision application as an optimal edge detection algorithm. Detailed comparison of parallel and sequential algorithm implementations is also presented.
Keywords :
computer graphic equipment; computer vision; coprocessors; edge detection; feature extraction; multi-threading; parallel algorithms; parallel architectures; performance evaluation; CUDA; Canny edge detector; Nvidia GPGPU; barrier synchronization; compute unified device architecture; computer vision application; feature extraction algorithm; image processing algorithm; multithreaded many core GPU; optimal edge detection algorithm; parallel algorithm; performance evaluation; sequential algorithm; shared memory; software platform; thread block; Graphics processing unit; Hardware; Image edge detection; Instruction sets; Kernel; Pixel; CUDA; GPU computing; Image Processing; Streaming Processor (SP); Streaming multiprocessor (SM); parallel computing;
Conference_Titel :
Communication Systems and Network Technologies (CSNT), 2011 International Conference on
Conference_Location :
Katra, Jammu
Print_ISBN :
978-1-4577-0543-4
Electronic_ISBN :
978-0-7695-4437-3
DOI :
10.1109/CSNT.2011.114
