Adaptive Gaussian mixture model based on feedback mechanism
Jinman Luo, Juan Zhu · 2010
Focusing on the traditional Gaussian mixture model suffers from slow learning and lack of accuracy, this paper proposes an adaptive Gaussian mixture model based on feedback mechanism. It models each pixel as an adaptive mixture of Gaussians, uses the information of foreground to advance model update based on feedback mechanism and selects the number of components of Gaussian mixture model adaptively to reduce convergence time of model update. Additionally, to improve model's ability of anti-disturbance and make it more accurately, a method of partial color similarity based on foreground matching is proposed. Experimental results demonstrate the algorithms our proposed is effective, low of algorithm complexity and robust.