Real-Time Object Detection with Simultaneous Denoising using Low-Rank and Total Variation Models

Abstract

The detection of objects in video scenes is the most prominent research topic in computer vision. It is the result of a wide variety of applications, such as virtual reality and intelligent surveillance systems, so that the system is based mainly on the detection of objects or moving objects. Due to the great success of the Foreground/Background Separation algorithms by decomposing the low-order matrix recently, we propose a new real-time incremental algorithm based on Low Rank and Total Variation (TV) model while simultaneously eliminating various noise. In this research, the proposed algorithms applied to solve multiple problems, such as Dynamic Background, Variation Background, with time. Also, we used real or online videos that will allow the adaptive modeling method to automatically remove noise and detect foreground (or intruder) on such scenes. The background modeling challenges in videos do not involve environmental differences such as lighting or weather changes. To check the effectiveness and efficiency of the proposed algorithms, we have experimented with real-time videos. Analytical experiments and results show the ability and efficiency of our method as well as the low computational cost of our proposed algorithms.