The Understanding of Meaningful Events in Gesture-Based Interaction - Intelligent Video Event Analysis and Understanding - page 5

equipments that are being held or worn and which provide additional levels of
accuracy and precision for the interaction. Hand tension or muscle activity rep-
resent such examples [16].
Figure 2 visually illustrates the four events in conjunction with their relationship
to various information categories they support. For example, location events can
identify where the action takes place, when it took place but also who performed it
by position reasoning and inferring. Posture events specify when a gesture started,
when it ended, whether it passed through some intermediate state as well as what
that gesture represented. Even more, executing a given posture is related to inten-
tion to interact which is clearly understood by both the system and user. Tap and
touch allow identification of where and when the action took place and they are
also strongly connected to intention (touching clearly specifies interest, designation
and willingness for interaction from the user's perspective). Custom events may be
specifically designed in order to address any of these questions.

Fig. 2 Event types in video processing (location, posture, tap, touch and custom events) and
their relationships to the information categories they address.

2.1 Location-Based Events

Location criteria can be used in order to define regions of interest in the video se-
quence to be processed where hands or other body parts are likely to be positioned
at the start, during or at the end of a gesture. The advantages are multiple:

First of all, the amount of image processing that is required is reduced only to
the most promising regions that will likely contain body parts engaged in gesture
production;

￿

The accuracy of the overall procedure is likely to become higher as the rest of
the image is simply ignored which considerably reduces the chance of outputting

￿