Texte intégral / Full text (pdf, 20 MiB) - Infoscience - EPFL
Texte intégral / Full text (pdf, 20 MiB) - Infoscience - EPFL
Texte intégral / Full text (pdf, 20 MiB) - Infoscience - EPFL
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Chapter 4. Simulating Visual Attention for Crowds<br />
putation is done with the different activation time values for the three sets of joints (eyes,<br />
head and torso). As depicted in Figure 4.3, we thus obtain a slight delay in the movement<br />
initiation between these three sets of joints.<br />
Temporal propagation fp(t)<br />
1<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0 10 <strong>20</strong> 30<br />
Frames<br />
Figure 4.3: Desynchronization between the eyes, head, and torso. The eyes start moving before<br />
the head and satisfy the constraint first. The head and cervicals start moving and satisfy the<br />
constraint before the remainder of the spine.<br />
Our final movement therefore allows the eyes to converge on the interest point and then<br />
partially re-center with respect to the head as the remainder of the joints move to satisfy the<br />
constraint. Indeed, they only partially re-center as a whole portion of the rotation is done<br />
by the eyes only. In our examples, most characters are in movement and the majority of<br />
the gaze constraints are associated to other characters in movement. These constraints are<br />
thus dynamic. We therefore recompute the displacement map to satisfy the constraint at each<br />
timestep. We can assume that the constraint’s position from one frame to the next does not<br />
change much. We therefore recompute the rotation to be done at each frame but maintain<br />
the total contribution fP (t)ci to apply which we calculated before the initiation of the gaze<br />
motion. However, we reset the contributions to 0 if the gaze constraint changes, i.e., if it<br />
is associated to another entity situated elsewhere in the scene. More specifically, this is the<br />
case when the current constraint location is farther than a pre-determined threshold from the<br />
constraint location at the previous frame. The newly calculated rotations to be performed<br />
by the joints to attain the new constraint’s position are then distributed over the appropriate<br />
number of frames.<br />
4.5 Results<br />
We used our framework to create some examples of the possibilities of our method. The motion<br />
clips for our examples have been sampled at 30 fps. All the animations were generated<br />
on an Intel Core 2 Duo 3.0 GHz, 2GB RAM, NVidia GeForce 8800 GT.<br />
58