Event Understanding of Human-Object Interaction: Object Movement Detection via Stable Changes - Intelligent Video Event Analysis and Understanding - page 206

4.3 Object Detection and Background Update

By using the detection result of the extracted foreground regions and removed-layer
regions, the method detects object placement by detecting static regions which the
input image's edge boundary energy is larger than the layered background's. In ad-
dition, the method detects object removal by detecting removed-layer regions and
static regions which the input image's edge boundary energy is smaller the layered
background's.
Then the method update the background model according to detection result.
Fig. 7 depicts an overview of background update. When the method detects placed
objects, the method inserts the detected objects to the layered background. When
the method detects removed objects by the layered background model, the method
deletes the detected objects from the layered background. When the method detects
removed objects by edge subtraction, the method updates the base background of
the detected object regions.

Placed-Object
Region

Removed-Object
Region
(By Layered Background)

Removed-Object
Region
(By Edge Subtraction)

Detection Result

Insert
Layered Background

Remove
Layered Background

Layered Background Layered Background

Update
Base Background

Base Background Base Background

Fig. 7 Overview of background update. The base background and the layered background
are updated according to object detection results

5R su s

5.1 Experiments

We evaluate the object movement detection performance of proposed method with
8 video sequences which are captured in 4 viewpoints (total 32 different video se-
quences, total 10324 frames). These video sequences consist of images of 320 240

resolution recorded at 7.5 fps. The video sequences contain 94 object placement and
58 object removal (total 152 events).

×