Acceleration of profile creation for three-dimensional vector video with GPGPU

Abstract

In the report the optimization of image similarity metric computation method for three dimensional vector video with general-purpose computations on graphical processor unit (GPGPU) is discussed. The use of stream processors in graphics accelerators and Compute Unified Device Architecture (CUDA) platform allows significant performance gain in comparison to calculations on general-purpose processors, while solving problems of computer vision and image similarity determination. The performance of the GPGPU metric value computation is measured and researched.

Keywords

• трехмерное видео, графический процессор, компьютерное зрение, метрика, ключевые точки