VNU-UET Repository: No conditions. Results ordered -Date Deposited. 2024-03-29T02:13:00ZEPrintshttp://eprints.uet.vnu.edu.vn/images/sitelogo.pnghttps://eprints.uet.vnu.edu.vn/eprints/2018-10-29T04:31:03Z2018-10-29T04:31:03Zhttp://eprints.uet.vnu.edu.vn/eprints/id/eprint/3123This item is in the repository with the URL: http://eprints.uet.vnu.edu.vn/eprints/id/eprint/31232018-10-29T04:31:03ZSelecting active frames for action recognition with 3D convolutional networkRecent applications of Convolutional Neural Networks, especially 3-Dimensional Convoltutional Neural Networks (3DCNNs) for human action recognition
(HAR) in videos have widely used. In this paper, we
use a multi-stream framework which is a combination
from separated networks with different kind of input
generated from unique video dataset. To achieve the
high results, firstly, we proposed a method to extract
the active frames (called Selected Active Frames -
SAF) from a videos to build datasets for 3DCNNs in
video classifying problem. Second, we deploy a new
approach called Vote fusion which considered as an
effective fusion method for ensembling multi-stream
networks. From the various datasets generated from
videos, we extract frames by our method and feed
into 3DCNNs for feature extraction, then we carry out
training and then fuse the results of softmax layers
of these streams. We evaluate the proposed methods
on solving action recognition problem. These method
are carried on three well-known datasets (HMFB51,
UCF101, and KTH). The results are also compared to
the state-of-the-art results to illustrate the efficiency
and effectiveness in our approachTieu Binh Hoangbinhhoangtieu@gmail.comThi Chau Machaumt@vnu.edu.vnSugimoto AkihiroThe Duy Buiduybt@vnu.edu.vn