Moving target detection method based on low-rank and sparse decomposition under dynamic background

Abstract

The invention belongs to the field of computer vision, and relates to a moving target detection method based on low rank and sparse decomposition under a dynamic background. According to the method, firstly, a gamma norm is introduced to approximate a rank function in an approximately unbiased manner so as to solve the problem that the obtained minimization problem cannot obtain the optimal solution and further reduce the detection performance due to the fact that a nuclear norm excessively punishes a large singular value, and the L1 / 2 norm is utilized to extract a sparse foreground target so as to enhance the robustness to noise. Based on the sparse and spatial discontinuity characteristics of the false alarm pixels, the spatial continuity constraints are proposed to suppress dynamic background pixels, and then a target detection model is constructed, and the obtained optimization problem is solved by using an augmented Lagrange multiplier method based on alternating direction minimization strategy extension. According to the invention, the moving target detection precision under the dynamic background condition is obviously improved.

Description

technical field [0001] The invention belongs to the field of computer vision and relates to a moving target detection method based on low-rank and sparse decomposition under a dynamic background. Background technique [0002] Moving object detection is one of the most dynamic research directions in the field of computer vision, and it has a wide range of applications in traffic monitoring, traffic flow detection, augmented reality, etc. As the first step in automatic video analysis, moving object detection aims to identify and segment objects of interest, thereby providing a basis for subsequent object tracking and behavior recognition. [0003] In recent years, many video-based moving target detection methods have been proposed one after another, and the related algorithms can be roughly divided into the following three categories: frame difference method, optical flow method and background subtraction (BS) method. Among them, the frame difference method is fast and simple...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problem that the accuracy of moving target detection is significantly reduced due to background motion, the present invention proposes a moving target detection method based on low-rank and sparse decomposition in a dynamic background

[0005] The basic idea of ​​realizing the present invention is, based on the RPCA detection model, first introduce the γ norm to approximate the matrix rank function to solve the biased estimation caused by the excessive punishment of the larger singular value of the nuclear norm, resulting in the inability to obtain the optimal solution for the minimization problem and thereby reducing the dynamics. The problem of target detection performance; then for L 1 Norm isolates the problem that each element does not fully use the spatial prior information of the foreground pixel, resulting in poor detection performance. The present invention uses L 1 / 2 The norm imposes sparse constraints on the foreground to suppress the false alarm phenomenon caused by the dynamic background; then, based on the sparseness and spatial discontinuity of the false alarm pixels, the spatial continuity constraints are imposed on the foreground; finally, using the Alternating Direction Minimization (ADM) The model obtained by solving the augmented Lagrangian multiplier (ALM) method of strategy expansion

Images