Motion Detection with Background Clutter Suppression Based on KDE Model

摘要

Background modeling is an important component of visual surveillance system. In some complicated outdoor system, such as traffic scene in night, solutions to problems as illumination and shadow disturbance are provided. The kernel density estimation is exploited to estimate the probability density function of background intensity and then to classify the pixel into background or foreground scene. Toward the modeling of dynamic characteristics, a normalized color space is proposed as part of a five-dimensional feature space. And experiment demonstrates the performance of the proposed approach.