Moving object detection based on smoothing three frame difference method fused with RPCA

摘要

In order to extract the human moving object more accurately and efficiently in the surveillance video, a moving object detection algorithm combining smoothing frame difference method and Robust Principal Component Analysis (RPCA) is proposed. In view of the "shadow" and "cavity" problems in the traditional three-frame difference method, each frame image converted into a gray image is first divided into a fuzzy set such as a smooth region, a texture region and an edge region, and the smooth region can reduce the sudden change of the light. The effect on the gray value, that is, the smoothing frame difference method; RPCA can achieve both data dimensionality reduction and high noise, spike noise rather than Gaussian distribution noise. The two algorithms are used in combination, and the background of the current frame of the RPCA extracted video is used as the intermediate frame of the smoothed frame difference method, and is respectively differentiated from the previous frame of the current frame and the current frame of the video, thereby avoiding the background pixel point. The influence eliminates the phenomenon of "cavity" and also contributes greatly to the reduction of noise. Video detection experiments in different scenarios show that it is more efficient and accurate than similar algorithms.