Event Understanding of Human-Object Interaction: Object Movement Detection via Stable Changes - Intelligent Video Event Analysis and Understanding - page 205

Fig. 6 Region-state update
by the state machine. The
attribute transfer from the
transition region to the
update-prohibited region
or the stable region when
the stability value is greater
than S th and less than − S th ,
respectively. The regions
classified as the update-
prohibited region never
transfer to another attribute.
The regions classified as
the stable region are deleted
from the foreground region
list and are inserted to the
layered background.

Update-prohibited
Region
(Non-object)

S(t) > S

th

Transition
Region

S(t) < - S th

Stable Region
(Object)

( C th ), the ratio of the major axis to the minor axis of the region by approximating the
region to ellipse ( E rth ) and the length of the minor axis of the region ( L mt h ), and the
average width of the region ( W th ). Bhattacharrya distance d h (

q ( E i ) , q ( E b )) between

a histogram q ( E i ) and a histogram q ( E b ) is calculated as follows.

1

k = 1 q ( E i ; k ) q ( E b ; k )

N

d h ( q ( E i ) , q ( E b )) = − (8)

As mentioned in Section 3.2, the regions where objects are placed and the regions
where initial objects are removed are classified into foreground state. To distinguish
this two regions, we employ a classification method based on edge subtraction [19].
Generally, the region where objects don't exist is less textured than where ob-
jects exist. So, textures of the region will increase when objects are placed, and
will decrease when objects are removed. The method extracts the amount of edge
energies in the boundary of the stable region of the input image and the layered
background, and if the energy of the input image is greater than the one of the lay-
ered background, the method classifies the region as object placement. Otherwise,
the method classifies the region as object removal. The method determines object
placement and object removal via the edge difference measure E c as defined below:

1
|

= r ∈ R c

| ∑ s ( || i r − i s ||

2

i r i s

2

E c −|| − || ) (9)

s

where R c is the contour of the extracted region, i r and i r are the RGB vector of
pixel r on the input image and those on the background image, respectively. s is
the 4-neighbor pixels of the pixel r .If E c > 0, the region is determined as object

placement, and if E c < 0, the region is determined as object removal.