An improved Gaussian mixture model for an adaptive background model

作者：Zhu, Qi Dan; Li, Ke^*; Zhang, Zhi; Cai, Cheng Tao

来源：Journal of Harbin Engineering University, 2010, 31(10): 1348-1353+1392.

DOI：10.3969/j.issn.1006-7043.2010.10.013

• Adaptive background model
• Adaptive learning rates
• Background model
• Classical algorithms
• Filtering method
• Gaussian Mixture Model
• Improved algorithm
• Improved methods
• Learning rates
• Moving targets
• Original algorithms
• Pixel values
• Short periods
• Speed-ups
• Surveillance video
• Visual effects
• Visual-processing

The Gaussian Mixture Model (GMM) has been widely used for modeling backgrounds. Aiming to overcome the defect in practical application, the classical algorithm was improved in two ways. The pixel filtering method and the new adaptive learning rate method were presented. The pixel filtering method recorded the pixel value of a point in a short period of time, and then analyzed this data. According to the pixel mean and variance, the dynamic interfering pixels of fast moving targets can be filtered out. The formation of the background was divided into four stages in the new adaptive learning rate method. Additionally, different stages were assigned different learning rates, which can speed up the background of the formation and regression. Compared with the original algorithm, the experimental results showed that the improved algorithm has a good visual effect in the surveillance video of two streets, and the background forms more quickly and clearly. The improved method, which can be applied to other visual processing algorithms, enhances the robustness and accelerates the formation of the background.

• 出版日期2010
• 单位哈尔滨工程大学