Event Understanding of Human-Object Interaction: Object Movement Detection via Stable Changes - Intelligent Video Event Analysis and Understanding - page 207

The object detection method was implemented on a PC with an Intel Core 2 Duo
2.5 GHz processor. The method ran with single-thread processing.
In this experiment, we implement false positive and recall as performance evalu-
ation measures, as defined below:

correctly detected object movements
total detected object movements

false positive = 1 − (10)

correctly detected object movements
total object movements in the images

=

recall (11)

In this experiment, we calculated performance of the proposed method under vari-
ant threshold parameters R th , C th , E rth , L mt h , W th . We compare the proposed method
with the proposed method without the layered background model (object movement
detection by only edge subtraction), and our previous method by the pixel-level lay-
ered background model [20]. The original object movement detection method via
the pixel-level layered background model cannot handle object movements which
exists in initial state, we added edge subtraction based object movement classifica-
tion (mentioned in 4.2) to the method.
Fig. 8 shows the resulting detection performance in various parameters. As can
be seen from the graph, where false positive is from 0.05 to 0.25, the detection
performance by the region-level background subtraction method is superior to the
detection performance by the pixel-level background subtraction method. This is
because of that the region-level background subtraction method is robust to back-
ground clutter (e.g. small object shift, small shadow regions).

1

0.8

0.6

layeredgraphcut
nonlayeredgraphcut
layeredpixellevel

0.4

0.2

0

0 0.1 0.2 0.3 0.4 0.5

false positive

Fig. 8 ROC curves of the proposed method. The solod line: by the proposed method. The
broken line: by the proposed method without the layered background model. The dashed-
dotted line: by our pixel-level object movement detection method [20].