Representing discrimination of video by a motion map

Wennan Yu, Yuchao Sun, Feiwu Yu, Xinxiao Wu*

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Representing the content of the video by a motion map is a challenging problem in video analysis. This paper proposes to integrate the discriminative information of a video into a map by optimizing the recognition accuracy of the original video in the action recognition task. The motion map represents a prefix of video frames sequence. A motion map and the next video frame can be integrated to a new motion map by the proposed 3-dimensional convolution based model. This model can be trained by incremental length clips from training videos iteratively, and the final acquired network can be used for generating the motion map of the whole video. Experimental results on the UCF101 and the HMDB51 datasets show that our method achieves better results compared with other related methods.

Original languageEnglish
Title of host publicationAdvances in Multimedia Information Processing – PCM 2017 - 18th Pacific-Rim Conference on Multimedia, Revised Selected Papers
EditorsBing Zeng, Hongliang Li, Abdulmotaleb El Saddik, Xiaopeng Fan, Shuqiang Jiang, Qingming Huang
PublisherSpringer Verlag
Pages703-711
Number of pages9
ISBN (Print)9783319773797
DOIs
Publication statusPublished - 2018
Event18th Pacific-Rim Conference on Multimedia, PCM 2017 - Harbin, China
Duration: 28 Sept 201729 Sept 2017

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume10735 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference18th Pacific-Rim Conference on Multimedia, PCM 2017
Country/TerritoryChina
CityHarbin
Period28/09/1729/09/17

Keywords

  • Action recognition
  • CNN
  • Discriminative information
  • Video analysis
  • Video-to-Image

Fingerprint

Dive into the research topics of 'Representing discrimination of video by a motion map'. Together they form a unique fingerprint.

Cite this