Event Understanding of Human-Object Interaction: Object Movement Detection via Stable Changes - Intelligent Video Event Analysis and Understanding - page 197

The structure of this chapter is organized as follows. Section 2 provides an
overview of our framework. Section 3 describes background modeling and our
tracking technique. Section 4 describes the classification method of extracted re-
gions for background updates and object detection. In Section 5, the experimental
results show that the proposed object movement detection method works well even
when human motions are present. Conclusions are presented in Section 6.

1.1 Related Works

To detect object movement, several types of approaches have been proposed. For ex-
ample, objects are considered as “highly featured points”, so attention point detec-
tion techniques are useful to directly detect objects in the images [4, 5]. Approaches
based of attention point techniques can be employed even with moving cameras.
But, the scene will be less featured when objects are removed, so these methods are
difficult especially to detect object removal.
As object detection methods based on background subtraction methods, the
method via detection of long-term changes [6] and methods via detection of back-
ground model adaptation caused by long-term changes [7, 8] has been proposed.
To reject changes caused by human regions, Maki employs a human detection ap-
proach based on the size of changed region, color pattern and face detection [6],
Harville employs human detection via height of changed regions and tracking [9]
and Grabner employs appearance-based human detection [10].

2 Overview of the Proposed Method

Fig. 2 depicts an overview of the proposed method. The proposed method has two
major stages: attentive region detection and object detection.
First, the method extracts changed regions by a background subtraction method,
and then tracks the extracted regions (“attentive region detection” stage in Fig. 2). In

Object Detection

Attentive Region Detection

(4) (5)

(1) (2) (3)

Pixel-level
State Detection

Region-state
Detection

Stable-change
Classification

Object
Detection

Blob
Detection

Region
Tracking

Image

(6)

Update regions' template
patches and
add new detected regions

Background
Update

Update regions'
attribute

Identify detected regions
by comparing with stored regions

Detect changed
pixels using
background models

Foreground Region List

Update background model
in static regions and
removed-layer regions

Layered Background Model

Fig. 2 Overview of the proposed method