Illumination change adaptive tracking based on color centroid shifting

Suk Ho Lee, Euncheol Choi, Moon Gi Kang

Research output: Contribution to journalArticle

4 Citations (Scopus)

Abstract

In this paper, we propose a stable illumination change adaptive tracking algorithm in which the color model update and the target localization are mutually constrained to each other. The mutual constraint is the result of sharing the same five-dimensional feature space, where the feature vector is composed of the three color components and the x,y coordinates of pixels inside the target region. The use of the five-dimensional feature vector introduces spatiality in the color model update, i.e., the reclustering of the feature space is constrained by the spatial location of the pixels corresponding to the target color. The spatial constraint on the color model update is also due to the use of a window which contains the pixels used in the color model update, where the location of the window is decided by the current target location. The color model update is performed by a five-dimensional mean shift-based clustering algorithm. The update in the color components in the five-dimensional feature vector handles the illumination change in the colors in the target region. The fivedimensional mean shift-based clustering updates also the x,y coordinates in the five-dimensional vector, and thus constrains the target localization. The target localization uses the x,y coordinates of the mean vectors of the five-dimensional vectors corresponding to each color bin. The target location is computed as a minimizer of a proposed energy functional, which is proposed such that the relative spatial locations of the mean vectors are considered in the target localization. The mutual constraint of the color model update and the target localization makes the tracking stable under various situations such as global illumination change, partial occlusion, and cluttered background. It can stably track the target even in the appearance of major background colors that are also major colors of the target object.

Original languageEnglish
Article number057205
JournalOptical Engineering
Volume50
Issue number5
DOIs
Publication statusPublished - 2011 May 1

Fingerprint

centroids
Lighting
illumination
Color
color
Pixels
pixels
occlusion
shift
Bins
Clustering algorithms

All Science Journal Classification (ASJC) codes

  • Atomic and Molecular Physics, and Optics
  • Engineering(all)

Cite this

@article{bb6cd333b8a44c5c8d5e3d25cbd356d1,
title = "Illumination change adaptive tracking based on color centroid shifting",
abstract = "In this paper, we propose a stable illumination change adaptive tracking algorithm in which the color model update and the target localization are mutually constrained to each other. The mutual constraint is the result of sharing the same five-dimensional feature space, where the feature vector is composed of the three color components and the x,y coordinates of pixels inside the target region. The use of the five-dimensional feature vector introduces spatiality in the color model update, i.e., the reclustering of the feature space is constrained by the spatial location of the pixels corresponding to the target color. The spatial constraint on the color model update is also due to the use of a window which contains the pixels used in the color model update, where the location of the window is decided by the current target location. The color model update is performed by a five-dimensional mean shift-based clustering algorithm. The update in the color components in the five-dimensional feature vector handles the illumination change in the colors in the target region. The fivedimensional mean shift-based clustering updates also the x,y coordinates in the five-dimensional vector, and thus constrains the target localization. The target localization uses the x,y coordinates of the mean vectors of the five-dimensional vectors corresponding to each color bin. The target location is computed as a minimizer of a proposed energy functional, which is proposed such that the relative spatial locations of the mean vectors are considered in the target localization. The mutual constraint of the color model update and the target localization makes the tracking stable under various situations such as global illumination change, partial occlusion, and cluttered background. It can stably track the target even in the appearance of major background colors that are also major colors of the target object.",
author = "Lee, {Suk Ho} and Euncheol Choi and Kang, {Moon Gi}",
year = "2011",
month = "5",
day = "1",
doi = "10.1117/1.3577662",
language = "English",
volume = "50",
journal = "Optical Engineering",
issn = "0091-3286",
publisher = "SPIE",
number = "5",

}

Illumination change adaptive tracking based on color centroid shifting. / Lee, Suk Ho; Choi, Euncheol; Kang, Moon Gi.

In: Optical Engineering, Vol. 50, No. 5, 057205, 01.05.2011.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Illumination change adaptive tracking based on color centroid shifting

AU - Lee, Suk Ho

AU - Choi, Euncheol

AU - Kang, Moon Gi

PY - 2011/5/1

Y1 - 2011/5/1

N2 - In this paper, we propose a stable illumination change adaptive tracking algorithm in which the color model update and the target localization are mutually constrained to each other. The mutual constraint is the result of sharing the same five-dimensional feature space, where the feature vector is composed of the three color components and the x,y coordinates of pixels inside the target region. The use of the five-dimensional feature vector introduces spatiality in the color model update, i.e., the reclustering of the feature space is constrained by the spatial location of the pixels corresponding to the target color. The spatial constraint on the color model update is also due to the use of a window which contains the pixels used in the color model update, where the location of the window is decided by the current target location. The color model update is performed by a five-dimensional mean shift-based clustering algorithm. The update in the color components in the five-dimensional feature vector handles the illumination change in the colors in the target region. The fivedimensional mean shift-based clustering updates also the x,y coordinates in the five-dimensional vector, and thus constrains the target localization. The target localization uses the x,y coordinates of the mean vectors of the five-dimensional vectors corresponding to each color bin. The target location is computed as a minimizer of a proposed energy functional, which is proposed such that the relative spatial locations of the mean vectors are considered in the target localization. The mutual constraint of the color model update and the target localization makes the tracking stable under various situations such as global illumination change, partial occlusion, and cluttered background. It can stably track the target even in the appearance of major background colors that are also major colors of the target object.

AB - In this paper, we propose a stable illumination change adaptive tracking algorithm in which the color model update and the target localization are mutually constrained to each other. The mutual constraint is the result of sharing the same five-dimensional feature space, where the feature vector is composed of the three color components and the x,y coordinates of pixels inside the target region. The use of the five-dimensional feature vector introduces spatiality in the color model update, i.e., the reclustering of the feature space is constrained by the spatial location of the pixels corresponding to the target color. The spatial constraint on the color model update is also due to the use of a window which contains the pixels used in the color model update, where the location of the window is decided by the current target location. The color model update is performed by a five-dimensional mean shift-based clustering algorithm. The update in the color components in the five-dimensional feature vector handles the illumination change in the colors in the target region. The fivedimensional mean shift-based clustering updates also the x,y coordinates in the five-dimensional vector, and thus constrains the target localization. The target localization uses the x,y coordinates of the mean vectors of the five-dimensional vectors corresponding to each color bin. The target location is computed as a minimizer of a proposed energy functional, which is proposed such that the relative spatial locations of the mean vectors are considered in the target localization. The mutual constraint of the color model update and the target localization makes the tracking stable under various situations such as global illumination change, partial occlusion, and cluttered background. It can stably track the target even in the appearance of major background colors that are also major colors of the target object.

UR - http://www.scopus.com/inward/record.url?scp=81755163689&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=81755163689&partnerID=8YFLogxK

U2 - 10.1117/1.3577662

DO - 10.1117/1.3577662

M3 - Article

AN - SCOPUS:81755163689

VL - 50

JO - Optical Engineering

JF - Optical Engineering

SN - 0091-3286

IS - 5

M1 - 057205

ER -