Background subtraction based on pixel clustering

Nowadays motion detection in video sequences is the basis for the high-level analytics and applications. After ViBe is proposed, the current mainstream detection algorithm is based on pixel level to construct the background model for the foreground / background segmentation. Influenced by this kind of algorithm, this paper proposes a spatiotemporal-based algorithm to construct an approximate precise background model. Difference with most of detection algorithms using the fixed threshold to judge the similarity of the pixel, this paper adopts the property of color clustering to determine the similarity dynamically. The experimental results on the 2012 Change Detection dataset show that the algorithm outperforms most state-of-the-art algorithms.

I. Introduction

With the development of society and technology, image processing and machine vision emerge in more and more areas. And the motion detection technology is a fundamental step for tracking, recognition and classification processing in video sequences analysis [1]. It is now widely applied in the field of mobile devices, military surveillance, medical research and video surveillance, etc.