Research On Moving Object Detection Algorithm Based On Low Rank And Sparse Decomposition

Moving object detection is one of the most basic tasks in computer vision.It plays a central role in object tracking,behavior recognition,scene understanding,video surveillance and other fields.The purpose of moving object detection is to identify the moving object in the image or video,and then extract the moving object from the background.With the increasing demand for high-quality video surveillance,researchers have developed various methods for moving object detection.However,due to the impact of numerous complicated elements,such as illumination change,shadow,dynamic background,camera jitter and noise,its accuracy and robustness still face great difficulties and challenges.The traditional robust principal component analysis approach is difficult to deal with multidimensional data,while the tensor principal component analysis fails to effectively utilize the low-rank prior information and spatio-temporal continuity of moving objects.To solve these problems,a moving object detection algorithm based on non-convex robust principal component analysis and total variational regularization is proposed.A moving object detection algorithm based on l_1/2 norm and total variational regularized low rank approximation.And a moving object detection method based on tensor low-rank approximation and tensor total variational regularization.The main contents and innovations of this paper are as follows:（1）Aiming at the complex factors such as illumination change,shadow and dynamic background in video sequences,a moving object detection algorithm based on non-convex robust principal component analysis and total variation regularization is proposed.Firstly,this paper introduces a non-convex function to deal with the issue that the nuclear norm severely penalizes large singular values for the sake of constraining the low-rank characteristics of video background availably.Then,the sparsity is strengthened with the l₁norm and the spatio-temporal continuity of the foreground is explored by total variational regularization.Finally,the augmented Lagrange multiplier algorithm is expanded by the alternating direction multiplier strategy to solve the model.The experimental results show that the proposed method is superior to the existing methods in terms of moving target detection accuracy and foreground extraction effect.（2）One of the major challenges that seriously affect the accuracy of moving object detection is the existence of noise in video sequences.Therefore,a new model via Robust Principal Component Analysis framework which utilizes l_1/2norm,l₂ norm and total variational regularization to detect moving objects from noisy video sequences is proposed.Firstly,the l_1/2 norm is introduced to enhance the sparsity of the foreground and the total variation regularization method is used to improve the spatio-temporal continuity of the foreground.Then,the l₂ norm regularization is used to detect and remove noise,which can simultaneously remove noise and detect moving objects.Finally,the augmented Lagrange multiplier algorithm is expanded by the alternating direction multiplier strategy to solve the model.Extensive experiments show that the proposed method outperforms existing methods in terms of the accuracy of moving object detection and the effect of foreground extraction from the video sequence with noise.（3）Traditional robust principal component analysis cannot deal with multi-dimensional data,and tensor principal component analysis fails to effectively utilize the low-rank prior information and spatio-temporal continuity of moving targets.A moving target detection algorithm based on non-convex tensor robust principal component analysis and tensor total variation is proposed.Firstly,the tensor non-convex function is used to solve the problem that the tensor nuclear norm cannot effectively solve the tensor rank.In addition,the l₁ norm is used to enhance the sparsity of the foreground,and the temporal and spatial continuity of the foreground is enhanced by tensor total variation,which fills the gap caused by the retained objects,so as to accurately extract the foreground.Finally,an efficient augmented Lagrange multiplier algorithm is used to solve the proposed algorithm model.The experimental results show that the proposed algorithm can not only process multi-dimensional data well,but also outperforms the existing methods in terms of moving target detection accuracy and foreground extraction effect.