Saturday, October 23, 2004

mapping activity to 3 states

input_activity-state_map.pdf

Let's say that these are the overlapping states as functions of activity:
Ghost state: when activity A or B close to 0,
else
Loner state: when activity A or B close to 1
Crowd state: when activity of A+ activity of B >> 1

Look at relative strength of A vs. B (measured not by ratio, but something more robust)
If A is more active, then perhaps we take motion data from A, and operate on live feed from B,
(and vice versa).

Use hysteresis (or simply smooth over time using line objects) to make state change stickier.

In the intermediate state, loner state, maybe we can cycle between (Yoichiro) fluid and (Delphine+Maria) smoke because for some parameter values, smoke can have watery response, and fluid can have smoky response (set heat diffusion high).
This cycling can depend on SOUND, perhaps, in which case jitter should READ (not just emit) some single parameter say in a circle-range [0,2pi] from Joels' code. In simple case, we can use elapsed clock (which in fact is already a function of accumulated total activity, so "time" essentially stops when no one is active the hall.)

At Van Nelle, we will tune the system so that 0 corresponds to residual light acivity when no one is moving, and 1-body is for 1 person's worth of "average" fidgeting. Dafualt parameters are already defined for the appropriate zmap's. (Maybe should use table, if we need a non-monotonic response curve.)

- xw

0 Comments:

Post a Comment

<< Home