Publication

Abstract : Advancements in high-quality video cameras and the consequent capture of minute details of the scene have led the field of computer vision to remarkable heights. This paper develops a tensor QR decomposition-based approach for Moving Object Detection (MOD), which aims to reduce the computational complexity without disturbing the structural framework of the input video frames. The increased performance and efficiency of the proposed method lie in the usage of tensor QR decomposition along with l 2 , 1 norm and l 1 / 2 norm. It is designed on top of a tensor-based Robust Principal Component Analysis (TRPCA) framework. In addition, this work safeguards the variation along the spatio-temporal directions with the effective use of Tensor Total Variation (TTV) regularization. The results and the analysis prove that the proposed method improves the F-measure by 15%–45% and reduces the computational complexity by 75%–85% with respect to the counterparts.