Improved Gaussian Mixture Models for Adaptive Foreground Segmentation

Nikolaos Katsarakis, Aristodemos Pnevmatikakis, Zheng-Hua Tan, Ramjee Prasad

Research output: Contribution to journalJournal articleResearchpeer-review

8 Citations (Scopus)

Abstract

Adaptive foreground segmentation is traditionally performed using Stauffer & Grimson’s algorithm that models every pixel of the frame by a mixture of Gaussian distributions with continuously adapted parameters. In this paper we provide an enhancement of the algorithm by adding two important dynamic elements to the baseline algorithm: The learning rate can change across space and time, while the Gaussian distributions can be merged together if they become similar due to their adaptation process. We quantify the importance of our enhancements and the effect of parameter tuning using an annotated outdoors sequence.
Original languageEnglish
JournalWireless Personal Communications
Volume87
Issue number3
Pages (from-to)629-643
Number of pages14
ISSN0929-6212
DOIs
Publication statusPublished - Apr 2016

Fingerprint

Gaussian distribution
Tuning
Pixels

Keywords

  • Adaptive foreground segmentation
  • Adaptive background mixture models
  • Gaussian Mixture Models

Cite this

Katsarakis, Nikolaos ; Pnevmatikakis, Aristodemos ; Tan, Zheng-Hua ; Prasad, Ramjee. / Improved Gaussian Mixture Models for Adaptive Foreground Segmentation. In: Wireless Personal Communications. 2016 ; Vol. 87, No. 3. pp. 629-643.
@article{bebada43bd2341c48390a9fa1cabdfb0,
title = "Improved Gaussian Mixture Models for Adaptive Foreground Segmentation",
abstract = "Adaptive foreground segmentation is traditionally performed using Stauffer & Grimson’s algorithm that models every pixel of the frame by a mixture of Gaussian distributions with continuously adapted parameters. In this paper we provide an enhancement of the algorithm by adding two important dynamic elements to the baseline algorithm: The learning rate can change across space and time, while the Gaussian distributions can be merged together if they become similar due to their adaptation process. We quantify the importance of our enhancements and the effect of parameter tuning using an annotated outdoors sequence.",
keywords = "Adaptive foreground segmentation, Adaptive background mixture models , Gaussian Mixture Models",
author = "Nikolaos Katsarakis and Aristodemos Pnevmatikakis and Zheng-Hua Tan and Ramjee Prasad",
year = "2016",
month = "4",
doi = "10.1007/s11277-015-2628-3",
language = "English",
volume = "87",
pages = "629--643",
journal = "Wireless Personal Communications",
issn = "0929-6212",
publisher = "Springer",
number = "3",

}

Improved Gaussian Mixture Models for Adaptive Foreground Segmentation. / Katsarakis, Nikolaos; Pnevmatikakis, Aristodemos; Tan, Zheng-Hua; Prasad, Ramjee.

In: Wireless Personal Communications, Vol. 87, No. 3, 04.2016, p. 629-643.

Research output: Contribution to journalJournal articleResearchpeer-review

TY - JOUR

T1 - Improved Gaussian Mixture Models for Adaptive Foreground Segmentation

AU - Katsarakis, Nikolaos

AU - Pnevmatikakis, Aristodemos

AU - Tan, Zheng-Hua

AU - Prasad, Ramjee

PY - 2016/4

Y1 - 2016/4

N2 - Adaptive foreground segmentation is traditionally performed using Stauffer & Grimson’s algorithm that models every pixel of the frame by a mixture of Gaussian distributions with continuously adapted parameters. In this paper we provide an enhancement of the algorithm by adding two important dynamic elements to the baseline algorithm: The learning rate can change across space and time, while the Gaussian distributions can be merged together if they become similar due to their adaptation process. We quantify the importance of our enhancements and the effect of parameter tuning using an annotated outdoors sequence.

AB - Adaptive foreground segmentation is traditionally performed using Stauffer & Grimson’s algorithm that models every pixel of the frame by a mixture of Gaussian distributions with continuously adapted parameters. In this paper we provide an enhancement of the algorithm by adding two important dynamic elements to the baseline algorithm: The learning rate can change across space and time, while the Gaussian distributions can be merged together if they become similar due to their adaptation process. We quantify the importance of our enhancements and the effect of parameter tuning using an annotated outdoors sequence.

KW - Adaptive foreground segmentation

KW - Adaptive background mixture models

KW - Gaussian Mixture Models

U2 - 10.1007/s11277-015-2628-3

DO - 10.1007/s11277-015-2628-3

M3 - Journal article

VL - 87

SP - 629

EP - 643

JO - Wireless Personal Communications

JF - Wireless Personal Communications

SN - 0929-6212

IS - 3

ER -