getting pitch, roll, yaw in exif or txt

[bboo]

I just joined ML and was looking through the form but didn't find any answer:

is there a way to get the camera's position (relative to itself = roll, pitch, yaw) and writing that to either EXIF data or some txt file on the card for every shot taken?
Would be very handy for greenbox photography, in order to match the correct perspective of the camera in CGI afterwards.

I guess the accelerometer/electr. level would constantly serve data and when the shutter is pressed one "sample" gets saved somewhere. so after the photoshoot I could have the correct axis for every dimension.

thanks a lot
keep it up

also: I am using a Canon 5Dmk3

[a1ex]

I've actually experimented with it (for the purpose of auto-rotating photos), but didn't publish the code yet.

Main issues:
- loses "sync" often (it reads bogus values until you take the battery out)
- camera seems to lock up (maybe I'm calling the routines at the wrong moments, not sure yet)
- the sensor is not very accurate (0.5 degree increments)
- if you move while taking the picture, the acceleration will influence your readings badly

=> I ended up fixing the rotation manually for a lot of pictures

If it wasn't for the instability issues, it would have been in the repository by now.

[jose_ugs]

5dm3 has an accelerometer??? r u sure???

[a1ex]

All Canon cameras that I've tested have such thing, but I don't know how to read it directly. Canon code exposes the values via a property that's pretty difficult to control and also unreliable.

Proof:
- for 5D3/60D/6D/7D press the INFO button
- for the others, check the threshold at which the photo orientation changes its value.

[jose_ugs]

now this is a good example how mind-blowing ML Video is... i haven't even fully realized what 5DM3 has hardware-wise...

[bboo]

I've actually experimented with it (for the purpose of auto-rotating photos), but didn't publish the code yet.

Main issues:
- loses "sync" often (it reads bogus values until you take the battery out)
- camera seems to lock up (maybe I'm calling the routines at the wrong moments, not sure yet)
- the sensor is not very accurate (0.5 degree increments)
- if you move while taking the picture, the acceleration will influence your readings badly

=> I ended up fixing the rotation manually for a lot of pictures

If it wasn't for the instability issues, it would have been in the repository by now.

looks like I'm better off getting all this data from the outside. I'm currently experimenting with a wii remote/IR LEDs, looks like the better option after your reply.

thanks a lot!