FIx issue with calculating iOS image orientation in certain special c…

[idealopamp]

…ases (#26786)

Uses accelerometer data to determine device orientation when the photo is taken. Fixes flutter/flutter#26786 .

[idealopamp]

One known issue with the current solution: You may notice that if you shake the device up/down in portrait or left/right in landscape that the orientation sometimes comes out wrong. I think this approach might be more robust against unusual amounts of acceleration, e.g. the user is in a bumpy car ride, if one were to take an average/smoothing of the accelerometer metrics over the past few seconds, but that requires a bit more work to implement, so just starting with this to see what you think.

[idealopamp]

One known issue with the current solution: You may notice that if you shake the device up/down in portrait or left/right in landscape that the orientation sometimes comes out wrong. I think this approach might be more robust against unusual amounts of acceleration, e.g. the user is in a bumpy car ride, if one were to take an average/smoothing of the accelerometer metrics over the past few seconds, but that requires a bit more work to implement, so just starting with this to see what you think.

[idealopamp]

One known issue with the current solution: You may notice that if you shake the device up/down in portrait or left/right in landscape that the orientation sometimes comes out wrong. I think this approach might be more robust against unusual amounts of acceleration, e.g. the user is in a bumpy car ride, if one were to take an average/smoothing of the accelerometer metrics over the past few seconds, but that requires a bit more work to implement, so just starting with this to see what you think.

Haven't written it in objective-c, but using the sensors library in dart, averaging over 200 milliseconds seemed to do pretty well at preventing jittery movements from affecting the outcome, although I did not test in an actual bumpy car.

[mklim]

Thanks again for investigating this and submitting a PR! I tested it out with front and rear cameras. It doesn't have any regressions and fixes the edge case that's currently broken now. This code LGTM pending a review from @cyanglaz to verify the objective c style.

[cyanglaz]

A tiny nit otherwise LGTM.