ENH: Idle sensor recalibration

[jschrempp]

When the Eyes code starts up it takes a reading from the 8x8 TOF sensor. This reading is then used as a background calibration. When sensor region values differ from this calibration frame we assume someone is standing there.

In operation the puppet head is often moved accidentally, or by a person who is looking into its inner workings. When this happens we need to reboot the Eyes code to get a new calibration frame.

With this enhancement I propose that after some period of time where the TOF regions have not changed, we take that frame as a new background calibration.

How long should this be? It could be rather short. The TOF sensor is very sensitive to changes (on the order of 1 mm); it is hard to stand in front of it without triggering a change. We will have to experiment, but it might be that 5 seconds without a change is enough to qualify for a new calibration frame.

What tolerance should we have to sensor region changes? If the sensor field is completely open, then changes are few. If the sensor field has some stationary objects in it, then we can get spurious value changes; we have attributed these to reflectance. We should examine what these spurious reports look like and compare them to the case when a human is actually in the field of view. It may be possible to determine these different cases. For example, spurious reports may tend to come in just one region at a time whereas it is hard for a person to disturb only one region.

[BobGlicksman]

The main issue that I see with this proposal is that the background calibration must be the "background". How can we be sure that this automatic recalibration is, in fact, sensing the background? What if the head is on a table at MN and someone moves a piece of equipment in front of it and then leaves the equipment there for a little while, e.g. to clear space to move the equipment into a permanent location. The calibration will see the equipment as background, then the equipment is moved and people coming into the TOF fov are no longer sensed because they are further away.

Here is an alternative proposal: we have pin A5 unused in the PCB. Perhaps we put a button on this pin -- press the button to recalibrate the TOF without the need to reboot the board? This way, we ensure that the head is again set up in the place where we want to calibrate.

[jschrempp]

I think I understand your point, but if the equipment was later moved away, then the auto-calibration would deal with that at the next auto-calibrate interval. That might be just five seconds.

[BobGlicksman]

It won't hurt to try the auto-recalibration and see how it works out. If we find operational problems, we can always back auto-recalibration out and put in a calibrate button.