Parallel implementation on a GPU of a detection and tracking algorithm by basic background subtraction

摘要

GPUs, have a highly parallel structure making them effective for a wide range of graphics tasks but also for processing tasks. We would like to exploit the highly parallel GPU structure to improve our computing performance for a detection and tracking algorithm by background subtraction with the CUDA programming language. In this paper, after the implementation of the basic motion detection algorithm on CPU, we implemented then our algorithm on NVDIA GPU to improve the execution time. A GPU profiling step is performed using the Nvdia Visual Profiler. Experimental results show a significant speedup around 161 compared to the CPU.