Adaptive Background-Updating and Target Detection in Motion State

A common method for moving target detection in image sequences involves a self-adaptive threshold for segmentation. This paper discusses each pixel modeling in the image as a mixture of Gaussian distributions and gives an improved method to update the model. First, the characteristics of the different frames method and a mixture of Gaussian distributions will be discussed. And then, the two methods, different frames method and a mixture of Gaussian distributions are combined to decide which learning rate is suitable for the updating method at this moment. If there is no moving target in the image sequences, the background model should be updated quickly to get the real background with low noise. The better foreground images will be gotten by this method with low computational complexity. The improved algorithm performs more robustly and powerfully than the classical Gaussian Mixture Model in moving target detecting.