Moving Object Detection Based On Subspace Learning

In computer vision,machine learning and pattern recognition,moving object de-tection has always been a popular research direction,and has received extensive attention of academia and industry.Moving object detection is mainly to detect the foreground moving objects in the scene by using the video sequences,including the walking pedestrians,driving vehicles,moving boats and so on.The moving object detection not only can be directly used in practical scene,but also can provide the ba-sis for video post-processing,including object recognition,object tracking,behavior analysis and so on.Therefore,moving object detection has important research signif-icance and practical value in intelligent monitoring and intelligent human-computer interaction and other applications.However,due to the video sequences from the real scene,there are many interference factors.Thus,moving object detection technol-ogy faces many challenges.On the other hand,subspace learning is a very popular research topic in recent years,which can quickly and accurately analyze the data by reducing the high-dimensional data to low-dimensional subspace.Therefore,the study of the subspace learning technology based moving object detection is mean-ingful both in theory and practical aspect.In this paper,the main research work is as follows:Firstly,we extensively investigate the relevant literatures of the existing moving ob-ject detection at home and abroad.And we briefly review the research progress in the field of moving object detection.After summarizing and classifying the exist-ing methods of moving object detection,we focus on the traditional moving object detection methods and moving object detection methods based on subspace learning.Secondly,we propose a matrix sparsity estimation method to estimate the sparse matrix quickly and accurately,and construct a low-rank decomposition space to decompose the low-rank matrix.Based on the low-rank decomposition space,a pro-jected gradient algorithm is proposed to solved the low-rank matrix quickly.The experimental results show that the moving object detection algorithm based on pro-jected gradient has a faster detection speed along with high detection accuracy.Finally,we adopt a novel rank estimation method to estimate the rank of low-rank matrix instead of the kernel norm used by the traditional method,and construct a new augmented lagrangian objective function.And we utilize the augmented la-grangian multiplier algorithm to solve the objective function.On this basis,we present the moving object detection method based on robust estimation and aug-mented lagrange multiplier.The experimental results show that the proposed algo-rithm has a good detection performance,especially whose detection speed is signifi-cantly faster than the current popular moving object detection algorithms.