Extended Gaussian Mixture Model Enhanced by Hole Filling Algorithm (GMMHF) Utilize GPU Acceleration

Abstract

Video surveillance system as one of the most important areas in computer vision needs background subtraction in their initial process. Recently there is no method that robust enough to handle all of the prevalent issues in this field, such as water rippling, waving trees, illuminations change, and slow motion object. In this study, we propose an enhanced Gaussian Mixture Model (GMM) with Hole Filling (HF) algorithm through the post processing operation. Graphic Processing Unit (GPU) acceleration is also applied to reduce the computational time. The experimental result shows that the proposed method can achieve 8x of speedup and give a better performance compared to the original extended Gaussian Mixture Model. It can remove most of the noises from various complex scenes of some benchmark publicly available data.