An efficient non-parametric background modeling technique with CUDA heterogeneous parallel architecture

Abstract

Foreground detection plays an important role in many content based video processing applications. To detect moving objects in a scene, the changes inherent to the background need to be modelled. In this work we propose a non-parametric statistical background modeling technique. Moreover, the proposed modeling framework is designed to utilize Nvidia’s CUDA architecture to accelerate the overall foreground detection process. We present three main contributions: (1) a novelty detection mechanism capable of building accurate statistical models for background pixels; (2) an adaptive mechanism for classifying pixels based on their respective statistical background model; and (3) the complete implementation of the proposed approach based on the Nvidia’s CUDA architecture. Comparisons and both qualitative and quantitative experimental results show that the proposed work achieves considerable accuracy in detecting foreground objects, while reaching orders of magnitude speed-up compared to traditional approaches.