Accurate Dynamic Scene Model for Moving Object Detection
Hong Yang, Yihua Tan, Jinwen Tian, Jian Liu · 2007
Adaptive pixel-wise Gaussian mixture model (GMM) is a popular method to model dynamic scenes viewed by a fixed camera. However, it is not a trivial problem for GMM to capture the accurate mean and variance of a complex pixel. This paper presents a two-layer Gaussian mixture model (TLGMM) of dynamic scenes for moving object detection. The first layer, namely real model, deals with gradually changing pixels specially; the second layer, called on-ready model, focuses on those pixels changing significantly and irregularly. TLGMM can represent dynamic scenes more accurately and effectively. Additionally, a long term and a short term variance are taken into account to alleviate the transparent problems faced by pixel-based methods.