基于稳健pca优化算法的视频背景建模分析.docx

Abstract
Video-based motion analysis mainly detect, track, classify, and recognize targets from the video sequences that contain a variety of moving targets, and all of the technology is based on the background modeling technique. Therefore, the study of background modeling has both important theory significance and application value.
Unlike traditional methods, the video background modeling based on Robust PCA can obtain stable and precise background without training step previously, and it is robust to the change of scene. This paper analyses the status and related application technologies of background modeling methods at first, and introduces the foundational theory of robust PCA. Then it focus on the subproblem about singular value position, the ALM (augmented Lagrange multiplier) algorithm and ADM (alternating direction method) algorithm for video background modeling, and improvement is carried out to achieve better results.
The putation of Robust PCA algorithms is the singular value position, and most algorithms only need to get the singular values that greater than a certain threshold. As plex structure and puting efficiency of PROPACK, this paper proposes threshold-based linear time singular value position by improving the linear time singular value position algorithm, which is simple to understand with high efficiency.
The ALM algorithm is very well for robust PCA, and it divided into exact ALM and inexact ALM. As the inexact ALM algorithm is better puting efficiency, the research bases on it in this paper, removing PROPACK package while introducing the threshold-based linear time singular value position. It can be found from the background modeling experimental results that the improved algorithm has better modeling result puting efficiency.
The ADM algorithm is also a good method for robust PCA, and it has good effect when doing experiments based on synthetic data. Thus, the paper makes it for the real video background modeling, and it can be found