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
Recommended Citation
Shi, Hang and Liu, Chengjun, "A New Global Foreground Modeling and Local Background Modeling Method for Video Analysis" (2018). Faculty Publications. 8989.
https://digitalcommons.njit.edu/fac_pubs/8989
