@InProceedings{BastosMeloSchw:2019:MuReRe,
author = "Bastos, Igor Leonardo Oliveira and Melo, Victor Hugo Cunha de and
Schwartz, William Robson",
affiliation = "{Universidade Federal de Minas Gerais} and {Universidade Federal
de Minas Gerais} and {Universidade Federal de Minas Gerais}",
title = "Multi-Loss Recurrent Residual Networks for Gesture Detection and
Recognition",
booktitle = "Proceedings...",
year = "2019",
editor = "Oliveira, Luciano Rebou{\c{c}}as de and Sarder, Pinaki and Lage,
Marcos and Sadlo, Filip",
organization = "Conference on Graphics, Patterns and Images, 32. (SIBGRAPI)",
publisher = "IEEE Computer Society",
address = "Los Alamitos",
keywords = "Gesture detection, gesture recognition, recurrent models,
multi-task.",
abstract = "Communication through gestures plays a relevant role in human
life, in which a non-verbal language is used to propagate
information among individuals. To recognize gestures, computers
need to represent and interpret human appearance and motion,
involving hands, arms, face, head and/or body, in a mathematical
sense. Despite the high applicability in different contexts, most
gesture recognition approaches in literature are not designed to
deal with unsegmented videos. That is, most approaches do not
temporally detect when a gesture occurs, which prevents to explore
correlations between detection and recognition tasks, besides
their application on real-world scenarios. In this sense, we
propose the Multi-Loss Recurrent Residual Network (MLRRN), a
multi-task based approach that performs both the recognition and
temporal detection of gestures at once. It employs a dual loss
function which takes into account the class assignment of each
frame of a video to a gesture class and also determines the frame
interval associated to each gesture. Our model counts with a dual
input, gathering information from appearance and human pose on
frames, besides bidirectional recurrent layers and residual
modules. According to experiments conducted on ChaLearn Montalbano
and ChaLearn ConGD datasets, our approach achieves results
comparable to state-of-the-art methods considering average
temporal Jaccard metric.",
conference-location = "Rio de Janeiro, RJ, Brazil",
conference-year = "28-31 Oct. 2019",
doi = "10.1109/SIBGRAPI.2019.00031",
url = "http://dx.doi.org/10.1109/SIBGRAPI.2019.00031",
language = "en",
ibi = "8JMKD3MGPEW34M/3U2KT7L",
url = "http://urlib.net/ibi/8JMKD3MGPEW34M/3U2KT7L",
targetfile = "camera_ready_mlrrn.pdf",
urlaccessdate = "2024, Apr. 28"
}