Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

Accelerometers in use

D

Daniel García

Well-known member
Joined
Oct 3, 2007
Messages
173
Reaction score
0
Points
0
One of the most salient newly announced DSMC features is the inclusion of 3-axis accelerometers and GPS sensors with each brain. The most obvious use for this rich metadata is as a substitute or maybe helper for traditional optical matchmoving processes for cgi integration. This is very nice forward thinking, but I wonder how new this concept is.
Given the availability and falling price of these kind of sensors I wonder if an external module exists that logs this data and is used for actual cameras nowadays. Maybe not as a commercial module, but as a propietary tool from the ILMs and DDs, used to help the turnaround on shows with hundreds of matchmoved shots.

I'm not sure if this method would provide for the sub-pixel accuracy normally needed for matchmoving, but it sure would help on ambiguous or hard to track shots.

My question is, does anybody know if this technique of camera matchmoving has been used before? any specific info on software, or gotchas?
 
Not just for Motion tracking...

http://www.scarletuser.com/showthread.php?p=116#post116


Camera shooting control
The user could set the camera to stop recording whenever it’s pointed at the ground, just in case the user has forgotten to turn off the cam.
It would also be possible to shut off shooting when the cam is wildly moving, and restart it when the cam has returned to a state where the recording isn’t going to make the audience puke (the Cloverfeild function.)
This could also be useful for quickly starting up a camera from being held in a relaxed position, to instant shooting when pointed at something interesting (usually stuff that is a bit more horizontal.)


‘Frame is level’ indication.
Say you’re shooting a wedding with your Scarlet cause the writers have gone on strike again, and you’re desperately trying to pay off a sub-prime mortgage.
Set the tripod down and start shooting cause the bride has just snogged the best man and you don’t want to miss the shot.
You have neglected to set the tripod level.
The Accelerometer notes that the cam is off horizontal and inserts the data into the metadata.
Back at ‘Weddings are Art Too ltd.’ you suck the clip up into AE (or what ever Post app) select the metadata plugin and the clip is instantly rotated to level.
 
I like the level. But I would go a step further and make it a UI function.

Create a virtual horizon line in the viewfinder. This would be great for handheld work if it was live on screen at all time. Provide a horizon line and a corresponding frame line for you to match.
 
Could warn you if your starting to drift.

Lie the cam on it's side and it's in 'standby.' Pick it up and it turns on.

Shut down the hard drives if there is too much vibration.
Shut down if the cam feels it's self falling (like the Lenovo laptops.)
 
I never dreamed that Red has been so fast of adapting inertial sensor technology in a camera. This is great news :) The most important feature for me for sure (except all new stuff we already knew)..

Daniel: The best inertial sensors have a 0,05 degree noise, that won't give you any "subpixel" accuracy for matchmoving shots. Also I don't think epic will reach a static accuracy below 1 degree in roll/heading/pitch. However, the great thing is that using inertial sensors together with matchmoving software gives you a very good accuracy in fast motion, when the motionblur makes the tracking software go nuts, or when you want to track in for example a dark environmet.. For example a strobo light scene, is impossible to matchmove, but with help of a sensor this might be very possible!

The sensor is also super when going from previz->live footage.. With the previz you know exactly the pitch/head/roll you want to set the camera at. With a built in sensor this information will be present to you at all time, making matched shots super easy!..

Big applause to Red for giving us this baaaad ass camera!
 
Exactly, Peter, great examples.

Now, my original question was... has this coupling of inertial + optical matchmoving been done before?
 
Yes.. But I only know of it as R&D tests... There is no off the shelf software to my knowledge that does this yet (Maybe because there aren't any cool cameras out there with built in sensors yet!? :)....)

I'll bet Andersson will be first out the door with a new version of Syntheyes :)....

Sofar Red has said very little about the /i Lens Data tech though... I still don't know why its not saved down in the metadata (they say its saved, but how the hell do you access it?!?!).... I hope this won't be the case with the inertial sensors...
 
So I've been pondering the 3d space in match-moving.
The cam is tracking it's movements, but we are tracking a relative position in space (relative to the last measurement taken or from when the record function is trigger for instance.)

What we need is an absolute position.
What if we had two radio sources such as two wifi phones positioned nearby.
The cam tracks it's position relative to these radio sources.

One could use the GPS to record spatial position?
 
You sure about that. I could have sworn I saw an option in PFTrack for importing tripod and head data from some of the newer remote head systems like mosys.

For instance on the Oconner 120EXe
The 120EXe provides output pan and tilt position information via an external 19 pin Fischer connector.
Mo-Sys provides a companion encoder box for the 120EXe which allows high resolution pan and tilt data of 1.8 million counts per revolution to be output from the head, making it suitable for film and HD formats.
When the 120EXe is coupled to the Mo-Sys Motion Logger, it instantly becomes a powerful camera motion capture system for use in post production. When used with the Mo-Sys 3D Inserter, it also allows real-time previsualisation of live action and CG compositing on set.
Click to expand...
 
DaveT said:
So I've been pondering the 3d space in match-moving.
The cam is tracking it's movements, but we are tracking a relative position in space (relative to the last measurement taken or from when the record function is trigger for instance.)

What we need is an absolute position.
What if we had two radio sources such as two wifi phones positioned nearby.
The cam tracks it's position relative to these radio sources.

One could use the GPS to record spatial position?
Click to expand...

I think D GPS does this but needs another data source and theres various very accurate surveying kit that could be used. Depends if other data input can be enabled.
 
I tend to agree that this is will be an aid to match moving, not a the end. These devices are good for measuring orientation, but far so less so at position data. So the camera is going to have a good idea of how it "turned" but not how far it moved relative to a given point. GPS would never work for this purpose (to the best of my knowledge) for many reasons. (line of sight, update rate, accuracy to name just a few)

Triangulation via other devices has it's own set of issues which are very difficult to achieve at the accuracy/update levels required.

So at the end of the day, at casual glance, this seems like it wil be a great aid. But will still require folks to buy and know how to use matchmoving software. No skipping it...yet anyway.
 
Even if it's not perfectly accurate - it could get you in the ball park which could be used to exclude some of the wilder matchmoving algorithm solutions and would make some matchmoving possible that previously was impossible.
 
All they need to do is, add Wii and iPhone capabilities and dominate the gaming and cellular phone markets! lol

Would be cool to have an XYZ coordinates for location shooting logs.
 
ericyoung said:
Even if it's not perfectly accurate - it could get you in the ball park which could be used to exclude some of the wilder matchmoving algorithm solutions and would make some matchmoving possible that previously was impossible.
Click to expand...

This is something that always sort of annoys me about matchmoving software. None of them take "hints" very well. PFTrack can sort of refine its track based on manual cues but none of them let the user give sufficient guidance IMO. It's often an all or nothing sort of situation. Hopefully with more sensor data being easily accessible and inexpensive we'll start to see some advancements on the software side related to 'rough assistance'.
 
You must log in or register to reply here.
Back
Top