Texte intégral / Full text (pdf, 20 MiB) - Infoscience - EPFL

Chapter 4. Simulating Visual Attention for Crowds
4.5.2 Gaze Behavior for Crowds
Figure 4.5: Examples of gaze attention behaviors in a crowd animation.
In this example, we illustrate the use of our scoring algorithm together with the motion
editing over 130 characters walking up and down a street, standing, or sitting on a bench.
This is depicted in Figure 4.5. The maximum distance threshold was set to 10 meters. The
attention threshold and the importance of each parameter was randomly generated by our
application and is different for each character. For each one, the scoring algorithm is applied
to all other eligible characters in the scene. Additionally, it is applied to all eligible scene
objects defined as potential interest points (60 in all). We can thus simulate a simple form
of top-down attention in the sense that some characters seem to be looking for something
or trying to find their way. An interesting aspect emerging from those results is that some
characters walking or standing next to each other regularly look at each other. We thus have
the impression that they are talking together.
4.5.3 Complexity and Computational Times
The complexity for the automatic interest point detection algorithm is in O(n 2 ) with n being
the number of characters. Indeed, for each character, we have to evaluate all other entities.
However, since we do not compute the interest points for entities out of the character’s field
60