Moving Object Detection Based on Truncated Nuclear Norm and 3D Total Variation

Abstract

Robust principal component analysis (RPCA) has been widely used in moving object detection, but the traditional low rank and sparse decomposition method constrained by nuclear norm can only be applied to simple scenes such as static background. In dynamic background, the traditional RPCA method cannot extract moving objects completely, and it is easy to detect the background as moving objects by mistake. To solve this problem, a moving object detection model based on truncated nuclear norm and total variation is proposed. The model uses truncated nuclear norm to better constrain the low rank of video background; Considering the temporal and spatial continuity of moving objects, the constraint of moving objects is realized by 3D total variation, so as to improve the accuracy of moving object detection. Finally, the two-step iterative strategy and augmented Lagrange multiplier method are introduced to solve the proposed model. Through simulation experiments and comparing the F-measure values of each algorithm, it is proved that the proposed model can effectively improve the accuracy of detecting moving objects in dynamic background from the perspective of vision and quantification, and achieved better visual effect than the existing model.