Spatio-temporal nonparametric background modeling and subtraction

Background modeling and subtraction is a core component of many vision based systems. By far the most popular background models are per-pixel models, in which each pixel is considered independently. Such models fail to handle dynamic backgrounds and noise. In this paper, we present a solution to this problem by proposing a novel and computationally simple spatio-temporal background model. We extend the nonparametric background model, one of the most widely used per-pixel models, from temporal domain to spatio-temporal domain. Instead of individual pixels, we consider 3 × 3 blocks centered on each pixel and use kernel density estimation (KDE) method in the 9-dimensional space. In order to reduce the computational complexity we use a hyperspherical kernel instead of Gaussian. We also make a small modification to the short term model used in order to handle sudden illumination changes. Experimental results show the effectiveness of the proposed model.