WIT Press


AUTOMATIC DETECTION OF LOITERING BEHAVIOUR USING SPATIOTEMPORAL IMAGE PROCESSING

Price

Free (open access)

Paper DOI

10.2495/CMEM190131

Volume

125

Pages

10

Page Range

133 - 142

Published

2019

Size

609 kb

Author(s)

YUTA EBIHARA, TERUOMI KATORI, TAKASHI IZUMI

Abstract

In this paper, the authors propose a method for detecting loitering behaviour automatically from security camera images acquired in a corridor or passage, and the authors examine the performance of the proposed method. Image sensors (security cameras) are widely used for crime prevention. In this study, for educational settings, the authors developed a system for automatically detecting loitering behaviour where a student is worried about whether he or she is permitted to enter a laboratory on his/her first visit. Using the results, staff in the laboratory can approach them and appropriately guide the student during his or her visit. The purpose of this study is to detect loitering behaviour including fuzzy actions. Detecting loitering behaviour involves the ethical issue of ensuring that the captured images do not infringe an individual’s privacy. In addition, there are a number of technical problems: What is a unique characteristic value indicating the target behaviour?; the method should not require much computational power; and it should be possible to explain the reason for the judgment result. In this study, to ensure privacy, the authors avoid using original images, for example, images in which the face or body of an individual can be recognized, and instead the authors use spatiotemporal images. General image processing is highly complex and requires computers using high-performance CPUs and a lot of memory. However, usual video capturing and behaviour recognition are expected to involve lower complexity. Spatiotemporal image processing can solve the technical problems mentioned above, for example, decreasing the computational complexity and maintaining high computational performance. In addition, as a measurement characteristic value, the authors adopt a simple staying time only, and the authors classify the behaviour into only two categories: “loitering behaviour” or “not loitering”.

Keywords

spatiotemporal image processing, loitering behaviour, staying time