A Research Method of Identifying a Speaker Based on Human Motion

DOI

10.2991/emcs-17.2017.64How to use a DOI?

Keywords

Speaker identification; Image processing; Face detection; Hand detection;Movement detection; F1 score

Abstract

In this paper, a novel method is purposed to identify a speaker in a video. Instead of using techniques of audio processing and lip movement, head motion and hand waving are adopted as a criterion to identify a speaker as those two kinds of movements are always accompanied when a person is speaking. It is obvious that a speaker is hard to identified accurately when the person is standing too far from the observing point to identify his/her lip movement or the surrounding is too noisy to identify his/her sound. Therefore, this method is purposed to identify a visual speaker based on the high level movement which avoids the two disadvantages mentioned before. Several different image processing algorithms are employed to detect the movement of face and hand. Moreover, the three-frame difference algorithm is modified to improve the accuracy and efficiency when detecting movements. Any other moving objects beyond the regions of face and hand will not be detected. F1 score is used to evaluate this system. After testing on 1,973 frames containing different occasions and characters, the average value of F1 score reaches 91.91% which proves the feasibility of this project. Furthermore, a conclusion can be drawn that the nearer the speaker is, the higher the F1 score is.

Copyright

© 2017, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Jiping Liu
AU  - Hong Gao
AU  - Chao Chen
PY  - 2017/03
DA  - 2017/03
TI  - A Research Method of Identifying a Speaker Based on Human Motion
BT  - Proceedings of the 2017 7th International Conference on Education, Management, Computer and Society (EMCS 2017)
PB  - Atlantis Press
SP  - 324
EP  - 329
SN  - 2352-538X
UR  - https://doi.org/10.2991/emcs-17.2017.64
DO  - 10.2991/emcs-17.2017.64
ID  - Liu2017/03
ER  -