(A) robust real-time head tracking algorithm based on color, shape, and quasi-spatial information칼라와 모양, 준공간 정보를 이용한 강인한 실시간 머리 추적 알고리즘

Cited 0 time in webofscience Cited 0 time in scopus
  • Hit : 503
  • Download : 0
In this paper, we propose a robust real-time head tracking algorithm using a pan-tilt-zoom camera. The proposed algorithm uses color, shape and quasi-spatial information for tracking. Based on the pre-obtained color histogram of a model, the proposed algorithm estimates the position of a target ellipse by maximizing the color histogram similarity (CHS) using the mean shift procedure, and adaptive gradient ascent method. And then it refines the scale so that the sum of gradient values near the boundary and the spatial color histogram similarity may be maximized. In the following frame, the initial position is selected at the same position of the ellipse in the previous frame. CHS is measured between the candidate color histogram and reference color histogram. Here, the reference histogram is obtained by combining the color histogram of the model and that of the tracking result in the previous frame. Thereby, the reference histogram partly reflects the recent change of the color of the target. The color histogram of the tracking result in the previous frame has not been usually used since the unwanted background color could be included into the reference color histogram due to the inaccurate positioning of the ellipse. To avoid this problem, we shrink the size of the ellipse of the previous frame adaptively to alleviate the unwanted error. For reliable tracking, an initial ellipse position should be near the target position. However, when the moving direction of the target changes drastically against that of the tracking camera, the background motion in the consecutive two frames is considerable. It often makes the initial position far from the target position; thereby the position may not converge to the target position. To alleviate the problem, we estimate a new initial position by compensating the background motion. To reduce the computational burden of motion estimation, we use vertical and horizontal 1-D projection datasets. In estimating the accurate posi...
Advisors
Ra, Jong-Beomresearcher나종범researcher
Description
한국과학기술원 : 전기및전자공학전공,
Publisher
한국과학기술원
Issue Date
2005
Identifier
244310/325007  / 020033577
Language
eng
Description

학위논문(석사) - 한국과학기술원 : 전기및전자공학전공, 2005.2, [ v, 56 p. ]

Keywords

adaptive previous histogram extraction; background motion compensation; region of convergence; Head tracking; spatial color histogram; 공간칼라히스토그램; 적응적 이전 히스토그램 추출; 배경움직임보상; 수렴영역; 머리추적

URI
http://hdl.handle.net/10203/37902
Link
http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=244310&flag=dissertation
Appears in Collection
EE-Theses_Master(석사논문)
Files in This Item
There are no files associated with this item.

qr_code

  • mendeley

    citeulike


rss_1.0 rss_2.0 atom_1.0