Every camera-driven crowd system rests on one assumption: the only thing changing in the picture is the audience. Automatic exposure destroys that assumption, and it does it in a way that is genuinely hard to diagnose from the front of house.
What actually happens
A camera on auto-exposure continuously adjusts to keep the overall image at a target brightness. Point it at an audience in a venue and the biggest brightness changes in its field of view are not the people — they are the lighting state.
When the LD pushes a look, the whole frame gets brighter or darker. The camera compensates. Every pixel changes value at once.
In movement mode, that reads as the entire room moving simultaneously — the strongest possible input signal, generated by nobody. In posture mode it is worse: the empty-room reference was captured at a different exposure, so the subtraction no longer isolates the people and the output becomes noise.
Why it is hard to spot
Because it correlates with the show.
Lighting changes tend to happen at exactly the moments the crowd also does something — a build, a drop, a reveal. So the system appears to respond, just erratically. Everyone concludes the tracking is unreliable, when in fact it is tracking perfectly and being fed a corrupted picture.
We lost most of a test session to this before working out that the game was following the lighting desk rather than the room.
What to lock
Exposure, gain, white balance and focus. Autofocus deserves its own mention: a camera that hunts for focus produces a whole-frame blur change, which reads exactly like motion.
If a camera cannot lock all four, it is not suitable, regardless of how good the image looks.
The consequence for lighting design
There has to be light on the audience, and it has to be stable for the duration.
That is a constraint on the lighting design, not a setting on our end, which means it belongs in a conversation with the LD during planning rather than a discovery at focus. Plenty of designs put everything into the stage and leave the house at near-black — a beautiful look that quietly makes camera-based crowd reading impossible.