Paper Abstract and Keywords |
Presentation |
2008-06-19 11:00
Real-time Anomal Movement Detection Based on Posision and Movement Direction of People Kyoko Sudo (NTT), Hiroyuki Ojima (Toyohashi Univ. of Tech.), Kazuyuki Iso, Kumi Jinzenji, Tatsuya Osawa, Hideki Koike, Hitoshi Nakazawa (NTT) DE2008-4 PRMU2008-22 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
We developed a method for detecting real-time anomal movement for
use in monitoring systems. Our method is based on learning the position and the moving
direction of people. The position and moving direction of people contains
important information that can be used for anomal detection; therefore, the tracks or the
flo histograms are used as features for anomal detection in conventional anomal detection
methods. However, it is difficult to use both the position information and the moving
direction information simultaneously because of the poor stability and the high calculation
cost. Therefore, we use the two types of information by extracting the
spatial and temporal average of the optical flow so that the time for the movement vector
continuously appears. We used an incremental SVM modified for 1-class as the discrimination
method, which enabled the online learning to be unsupervised. Experiments showed that when
using our method with QVGA-size video as the input, anomal scenes can be detected and that
real-time computation is possible. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
security video / anomal detection / optical flow / / / / / |
Reference Info. |
IEICE Tech. Rep., vol. 108, no. 94, PRMU2008-22, pp. 19-24, June 2008. |
Paper # |
PRMU2008-22 |
Date of Issue |
2008-06-12 (DE, PRMU) |
ISSN |
Print edition: ISSN 0913-5685 Online edition: ISSN 2432-6380 |
Copyright and reproduction |
All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
Download PDF |
DE2008-4 PRMU2008-22 |
|