Application of Pixel Drift Denoising Algorithm in Optimizing Gaussian Mixture Model

In real life, it is often necessary to detect moving targets in video streams captured by fixed video camera heads, and one of the key tasks is dynamic background modeling. Gaussian mixture model is a common modeling method. However, this modeling method is prone to random noise interference after extracting moving targets. How to eliminate or weaken such random noise is a research focus. Usually, there are two kinds of noise sources, that is, internal and external factors of video camera head imaging. The internal factors are related to manufacturing technology and materials of video camera heads, while the external factors are caused by micro-vibration, air disturbance and other factors. This paper focuses on how to eliminate or weaken interference of external factors. Firstly, a pixel drift denoising algorithm is proposed by analyzing formation mechanism of random noise caused by various external factors, that is, phenomenon of small scale drift of pixel position during imaging. Then, the pixel drift denoising algorithm is applied to Gaussian mixture model to determine foreground pixels, reduce noise impact, and improve integrity of moving targets. A comparative experiment is carried out in public data set CDnet 2014. The results show that in the same data set scene, the improved Gaussian mixture model algorithm integrating the pixel drift denoising algorithm can effectively reduce the noise in dynamic background, and the peak signal-to-noise ratio of experimental background and real background reaches 38. 2dB.