Real-Time Background Generation and Update Using Online CCTV Videos

Abstract

This paper presents a robust background image generation and update. The foreground (i.e. moving objects) is typically identified by the difference between two consecutive images or between a background (i.e., non-moving parts) and an input image. The former can detect moving objects faster while the latter can detect more accurately. The accuracy of the background image is very important in detecting the foreground image, especially in the background subtraction method. The speed of background generation and update is also important in real-time based operation. The proposed method can generate the background image well enough to detect the foreground objects and update the image fast enough to reflect the environmental changes in illumination and weather and scene changes due to camera motion. The proposed method includes two steps, the background generation and the background update. In the background generation step, a differential image of two consecutive images was created to separate moving and non-moving parts. That differential image was compared with the previous background image to generate an updated one. The experimental results using online CCTV videos at several different locations and traffic situations show that the method can perform robustly and effectively.