Automatic background extraction from video sequences

Abstract

Detecting and tracking moving objects is a basic and important task in many computer vision and video analysis applications such as video surveillance and object tracking. A background image without moving objects is needed to be used as reference information for further analysis. However, it is not possible to obtain a sequence of images without moving objects in, for example, traffic monitoring. The problem with background generation is one with great complexity. Many techniques have been derived in attempts to provide a solution. One such technique that aims to simplify the model shall be discussed and analysed here. This project implements an unconventional method of automatically generating a background from a given image sequence. It is prototyped in MATLAB and implemented using Microsoft Visual C++ and the OpenCV library.