@InProceedings{SilvaZavaBellSilv:2016:TrFaBa,
author = "Silva, Luan Porf{\'{\i}}rio e and Zavan, Fl{\'a}vio Henrique de
Bittencourt and Bellon, Olga Regina Pereira and Silva, Luciano",
affiliation = "{Universidade Federal do Paran{\'a}} and {Universidade Federal do
Paran{\'a}} and {Universidade Federal do Paran{\'a}} and
{Universidade Federal do Paran{\'a}}",
title = "Follow that nose: tracking faces based on the nose region and
image quality feedback",
booktitle = "Proceedings...",
year = "2016",
editor = "Aliaga, Daniel G. and Davis, Larry S. and Farias, Ricardo C. and
Fernandes, Leandro A. F. and Gibson, Stuart J. and Giraldi, Gilson
A. and Gois, Jo{\~a}o Paulo and Maciel, Anderson and Menotti,
David and Miranda, Paulo A. V. and Musse, Soraia and Namikawa,
Laercio and Pamplona, Mauricio and Papa, Jo{\~a}o Paulo and
Santos, Jefersson dos and Schwartz, William Robson and Thomaz,
Carlos E.",
organization = "Conference on Graphics, Patterns and Images, 29. (SIBGRAPI)",
publisher = "Sociedade Brasileira de Computa{\c{c}}{\~a}o",
address = "Porto Alegre",
keywords = "Face tracker, Nose region, Face image quality.",
abstract = "Face tracking uses temporal information to infer the position of
the face in each frame. One of its applications is in
unconstrained (in-the-wild) environments where face detection
methods fail to perform robustly. Current approaches presented in
the literature are based on facial landmarks. Therefore, they have
limitations when applied in in-the-wild environments as estimating
the landmarks in such scenarios is not trivial. To address this
issue, we propose a novel landmark-free approach based on a
state-of-the-art generic visual tracking method, as baseline,
combined with face quality assessment for initializing the
tracking. In addition, we introduce using only the nose region as
a solution for in-the-wild face tracking, initializing it with the
nose of the best quality face in the video sequence. The nose is
detected and used to estimate the head pose, which is combined
with the face quality score for choosing the initialization frame.
The nose region, rather than the entire face was chosen due to it
being unlikely to be occluded, mostly invariant to facial
expressions and visible in a long range of head poses. We
performed experiments on the 300 Videos in the Wild dataset and
our results favorably compared against the baseline method.",
conference-location = "S{\~a}o Jos{\'e} dos Campos, SP, Brazil",
conference-year = "4-7 Oct. 2016",
language = "en",
ibi = "8JMKD3MGPAW/3ME7PA8",
url = "http://urlib.net/ibi/8JMKD3MGPAW/3ME7PA8",
targetfile = "Follow_Sibgrapi_Noses.pdf",
urlaccessdate = "2024, May 03"
}