Real-Time Analysis of Human Body Parts and Gesture-Activity Recognition in 3D - 3D Modeling and Animation: Synthesis and Analysis Techniques for the Human Body - page 156

Figure 12. Recognized activity: Moving left.
Figure 13. Moving down and moving up activities.
Figure 12 shows recognized moving left activity. Figure 13, first row, shows
moving down activity where the person turns from the first camera to the second
one during the activity period. The second row is another correctly recognized
moving down/up activity for a different person.
Figure 14 is an example for unattended object detection. The person enters the
scene with an object in his hand and leaves the object on the table. After a
predefined time, the system can detect the body parts and the object left on the
table correctly. An alarm can be generated and sent to the security personnel for
the unattended object.
For applications such as smart rooms where devices are controlled by people, the
deployment of cameras can be adjusted for optimum capture of the motion.
However, for less structured motion estimation applications such as surveillance,
self-occlusion may occur. In this case, the corresponding parts are modeled less
accurately for the duration of the occlusion. One of our future works includes a
feedback scheme that uses temporal dependency. Furthermore, more cameras