A New Global Foreground Modeling and Local Background Modeling Method for Video Analysis

Document Type

Conference Proceeding

Publication Date

1-1-2018

Abstract

This paper presents a new Global Foreground Modeling (GFM) and Local Background Modeling (LBM) method for video analysis. First, a novel feature vector, which integrates the RGB values, the horizontal and vertical Haar wavelet features, and the temporal difference features of a pixel, enhances the discriminatory power due to its increased dimensionality. Second, the local background modeling process chooses the most significant single Gaussian density to model the background locally for each pixel according to the weights learned for the Gaussian mixture model. Third, an innovative global foreground modeling method, which applies the Bayes decision rule, models the foreground pixels globally. The GFM method thus is able to achieve improved foreground detection accuracy and capable of detecting stopped moving objects. Experimental results using the New Jersey Department of Transportation (NJDOT) traffic video sequences show that the proposed method achieves better video analysis results than the popular background subtraction methods.

Identifier

85050565460 (Scopus)

ISBN

[9783319961354]

Publication Title

Lecture Notes in Computer Science Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics

External Full Text Location

https://doi.org/10.1007/978-3-319-96136-1_5

e-ISSN

16113349

ISSN

03029743

First Page

49

Last Page

63

Volume

10934 LNAI

Grant

1647170

Fund Ref

National Sleep Foundation

This document is currently not available here.

Share

COinS