Deep learning for action recognition and prediction

Qiuhong Ke

Research output: ThesisDoctoral Thesis

300 Downloads (Pure)

Abstract

This thesis proposes novel deep leaning solutions for real-world applications, including action recognition and prediction from videos. The first part leverages Convolutional Neural Networks (CNN) to learn deep spatial-temporal information of skeleton sequences for action recognition. The second part presents three new methods based on CNN and Recurrent Neural Networks for action prediction. The last part of this thesis presents a new feature extraction method based on CNN for better person re­identification. Extensive experiments have shown the superiority of the proposed methods compared to state of-the-art methods for action recognition, action prediction and person re-identification.
Original languageEnglish
QualificationDoctor of Philosophy
Awarding Institution
  • The University of Western Australia
Thesis sponsors
Award date25 Jul 2018
DOIs
Publication statusUnpublished - 2018

Fingerprint

Neural networks
Recurrent neural networks
Feature extraction
Deep learning
Experiments

Cite this

@phdthesis{a5d9d5b049564d3a9a73b23e6779d050,
title = "Deep learning for action recognition and prediction",
abstract = "This thesis proposes novel deep leaning solutions for real-world applications, including action recognition and prediction from videos. The first part leverages Convolutional Neural Networks (CNN) to learn deep spatial-temporal information of skeleton sequences for action recognition. The second part presents three new methods based on CNN and Recurrent Neural Networks for action prediction. The last part of this thesis presents a new feature extraction method based on CNN for better person re­identification. Extensive experiments have shown the superiority of the proposed methods compared to state of-the-art methods for action recognition, action prediction and person re-identification.",
keywords = "deep learning, action prediction, convolutional neural networks, 30 skeleton sequences, recurrent neural networks, person re-identification, action recognition",
author = "Qiuhong Ke",
year = "2018",
doi = "10.26182/5b74d8e4281d1",
language = "English",
school = "The University of Western Australia",

}

Ke, Q 2018, 'Deep learning for action recognition and prediction', Doctor of Philosophy, The University of Western Australia. https://doi.org/10.26182/5b74d8e4281d1

Deep learning for action recognition and prediction. / Ke, Qiuhong.

2018.

Research output: ThesisDoctoral Thesis

TY - THES

T1 - Deep learning for action recognition and prediction

AU - Ke, Qiuhong

PY - 2018

Y1 - 2018

N2 - This thesis proposes novel deep leaning solutions for real-world applications, including action recognition and prediction from videos. The first part leverages Convolutional Neural Networks (CNN) to learn deep spatial-temporal information of skeleton sequences for action recognition. The second part presents three new methods based on CNN and Recurrent Neural Networks for action prediction. The last part of this thesis presents a new feature extraction method based on CNN for better person re­identification. Extensive experiments have shown the superiority of the proposed methods compared to state of-the-art methods for action recognition, action prediction and person re-identification.

AB - This thesis proposes novel deep leaning solutions for real-world applications, including action recognition and prediction from videos. The first part leverages Convolutional Neural Networks (CNN) to learn deep spatial-temporal information of skeleton sequences for action recognition. The second part presents three new methods based on CNN and Recurrent Neural Networks for action prediction. The last part of this thesis presents a new feature extraction method based on CNN for better person re­identification. Extensive experiments have shown the superiority of the proposed methods compared to state of-the-art methods for action recognition, action prediction and person re-identification.

KW - deep learning

KW - action prediction

KW - convolutional neural networks

KW - 30 skeleton sequences

KW - recurrent neural networks

KW - person re-identification

KW - action recognition

U2 - 10.26182/5b74d8e4281d1

DO - 10.26182/5b74d8e4281d1

M3 - Doctoral Thesis

ER -