A Background Modeling Algorithm Based on Improved Adaptive Mixture Gaussian

For better background modeling in scenes with nonstationary background, a background modeling algorithm based on adaptive parameter adjustment of the Mixture Gaussian is proposed. Mixture Gaussians is applied to learn the distribution of per-pixel in the temporal domain and to control the adaptive adjustment of number K of Gaussian components through increasing, deleting or merging similar Gaussian components adaptively. The new parameters C k and φ k are introduced in the adaptive parameter model. According to the actual situation, the adaptive adjustment of ρ can accurate track the real-time changes with the pixel, which improves the robustness and convergence. Experimental results show that the algorithm can rapidly response when the scene changes in the sequence of video with many uncertain factors, and realize adaptive background modeling with accurate target detection.