Berryrieser's analysis misses a few things.
- It's not necessary to use GPS to determine latitude since the firmware could just have an option to enter it (no different than the focal length of a manual lens, really). Sure, most people will forget to set it and never change it but it's still less dumb than assuming no rotation. Shipping cameras set to ±45 degrees is an obvious default, would correct reasonably well for most of the Earth's human population, and be better than zero for basically everyone else. Berryrieser's claim the variant he describes would not work without GPS is in the right direction but he misses some trivial solutions, such as continuing with the last known latitude when satellite signals are lost.
- The claim the camera must be pointed at the subject for an unrealistically long time is potentially true but does not appear to be generally supported. Since the beginning of significant gyro drift is 6.5 stops from the exposure that same amount of time is about what's needed to start up the highpass filter described. For example, 6.5 stops from 1/200 is 400 ms. Even in most quick shooting situations a camera would usually have another second or two to acquire gryo drift information. There's also no requirement drift be calibrated only during subject pointing, meaning potentially the entire uptime of the camera ever can be used. In practice, times where the camera's rotating too quickly would probably need to be excluded but acquiring calibration data whenever the camera's on really shouldn't be problem.
- The comments mention offset bias noise in gyros exceeding the Earth rotation component but don't complete the thought to using other sensors to define the zero drift point. In this case I'm not familiar with the specific sensors but usually there's some algorithmic iteration between different data sources to get best results. Precession is just another rotation matrix to include there. It does add another degree of freedom and make finding an optimal solution a bit more complex but I'm skeptical of describing it as a barrier.
Since automatic drift cal is the same thing as latitude learning it seems like it would remove need for GPS or latitude entry.
With regards to using GPS, yes, you are of course correct. You could have the user set the latitude, use the last known GPS, and the system would work most of the time. But if you get on a plane, land in a new place, and start shooting, your camera could be trying to correct for the wrong drift and mess the image up more than if we had no Earth rotation compensation. In this case I think the engineers would play it safe to avoid the risk of messing up some pro's shoot.
With regards to drift, you can't acquire drift information while handholding. The slow drift in your pointing will be greater than Earth's rotation rate. If you try to zero out the drift in this time, you are just going to be erroneously cancelling the drift rate from your hands which is going to be completely different a couple seconds later. For the is to work, you must be able be able to hold the camera steady with a DC drift rate of less than 0.004 degrees/sec (Earth's rotation rate). You could of course hold it steady for longer to integrate, but it would have to be long enough to make Earth's rotation large compared to your personal drift rate. I could probably hold a camera pointed at the subject to with a couple degrees over time. That would still require minutes of holding steady before the shot to effective measure and cancel Earth's rotation rate.
In order to calibrate away drift, the camera must be held steady. Preferably resting on a solid surface. This could measure and cancel Earth's rotation fairly quickly, but you can't set and forget. If you turn around 180 degrees, the effect from Earth's rotation will be in the opposite direction. You could of course have the gyro try to measure the camera's position at all times to do proper inertial sensing, but this would require your gyro to be running at all times, draining the battery. You might be able to pull in a magnetometer to help with orientation, but this is subject to its own pitfalls, particularly since the camera body is made of metal and does some shielding. Anyways, I certainly don't think it is impossible to compensate for Earth's rotation and is quite easy to do in a lab. It's just hard on a consumer camera.