FS7: Actual Active Sensor Area Size

V

Viddovation

Veteran
I'm trying to do some VFX testing with the FS7 that involves mapping the camera position in After Effects. Getting the right numbers on the sensor size and the lens focal length becomes exceedingly important. I'm trying to work it out with practical tests, but it's frustrating - prime lenses aren't necessarily exactly the focal length they claim, and I'm getting inconsistent results.


For the 3840x2160 16:9 active area, I'm getting results that suggest only 17-18mm of horizontal sensor area is active. (Which would come to more like 19.2mm for the 4096 horizontal area).

Now, I'm seeing posted numbers of the sensor being something like 23.6mm x 13.3mm, with 11.6mp photosites. But even if accurate, that would be the total size, not the active size... In 3840x2160 you'd only have 8.3mp active.

Anycase, I suck at all this math, and if anyone has done vfx with the FS7 or otherwise has a clue, definitely would love to hear from you.
 
Here’s my thought, and it may not be a very good one. Go into a dark room and set up a point light source on the other side of the room. Take off the lens. Use a piece of cardboard to mask off portions of the sensor. Watch a monitor. You should be able to determine the active image area, or at least get a very good approximation.
 
Viddovation said:
I'm trying to do some VFX testing with the FS7 that involves mapping the camera position in After Effects. Getting the right numbers on the sensor size and the lens focal length becomes exceedingly important. I'm trying to work it out with practical tests, but it's frustrating - prime lenses aren't necessarily exactly the focal length they claim, and I'm getting inconsistent results.


For the 3840x2160 16:9 active area, I'm getting results that suggest only 17-18mm of horizontal sensor area is active. (Which would come to more like 19.2mm for the 4096 horizontal area).

Now, I'm seeing posted numbers of the sensor being something like 23.6mm x 13.3mm, with 11.6mp photosites. But even if accurate, that would be the total size, not the active size... In 3840x2160 you'd only have 8.3mp active.

Anycase, I suck at all this math, and if anyone has done vfx with the FS7 or otherwise has a clue, definitely would love to hear from you.
Click to expand...

What you're doing is right, but i don't know how accurate AE is for anything other than superficial stuff. I assume you're camera tracking? I've only looked at AE a while back and it's one of those cases that it just either works, or fails dismally with little recourse.

I've been able to do camera tracking successfully in Nuke and PFTrack using the FS700 sensor size, 23.6 and 13.3. And i've done similar to you with an A7sII by working out the percentage crop based on the 3840 horizontal out of the total sensor width.

But in most cases the software will not need 100% absolute numbers and should iterate around until it finds a solution. Tracking is an iterative process of testing, removing trackers, testing, solving etc,. More often than not with tracking it's lens distortion that needs to be carefully corrected and sometimes manual tracking of points along with supplied survey of features (like distances between points in the real world). Most solutions work around all of these criteria to find a solve that works.

For the FS7, what physical dimensions are you getting for the sensor size? The only reference i can find is the same as above 23.6 x 13.3. If the full width of the sensor is 4096, then that 3840 would be horizontal of 22.125mm. I assume it's a centre crop. However most sensors include edge pixels for black balancing so there might be more than 4096 but in my experience you never need to get that anal about it.

The film back and focal length are connected together. As you change on the other changes too for the same given solve.

Does that help at all?

cheers
Paul
 
paulcurtis said:
What you're doing is right, but i don't know how accurate AE is for anything other than superficial stuff. I assume you're camera tracking? I've only looked at AE a while back and it's one of those cases that it just either works, or fails dismally with little recourse.

I've been able to do camera tracking successfully in Nuke and PFTrack using the FS700 sensor size, 23.6 and 13.3. And i've done similar to you with an A7sII by working out the percentage crop based on the 3840 horizontal out of the total sensor width.

But in most cases the software will not need 100% absolute numbers and should iterate around until it finds a solution. Tracking is an iterative process of testing, removing trackers, testing, solving etc,. More often than not with tracking it's lens distortion that needs to be carefully corrected and sometimes manual tracking of points along with supplied survey of features (like distances between points in the real world). Most solutions work around all of these criteria to find a solve that works.

For the FS7, what physical dimensions are you getting for the sensor size? The only reference i can find is the same as above 23.6 x 13.3. If the full width of the sensor is 4096, then that 3840 would be horizontal of 22.125mm. I assume it's a centre crop. However most sensors include edge pixels for black balancing so there might be more than 4096 but in my experience you never need to get that anal about it.

The film back and focal length are connected together. As you change on the other changes too for the same given solve.

Does that help at all?

cheers
Paul
Click to expand...

Paul,

I'm not doing standard tracking points in software.

What I have is a slide/pan/tilt Kessler Second Shooter rig for my real world camera, and I have custom software that executes programmed moves, and saves motor encoder position data for every video frame. Alternatively, it generates "virtual" reports, which would be the expected motor encoder positions for each video frame. I then take either report, and convert it into After Effects units - pixels (at 72 pixels per real world inch) and degrees. The math is tedious, but I've got it right at this point. I use a series of null points to build a skeleton of the slider, with the After Effects camera mounted at the top.

The upside of this approach is that it can work on shots that don't have viable tracking points for post software to calculate from.

I'm realizing that some of my issue is that the lens I've been using for testing breathe too much, and despite being primes, have different focal lengths depending on where they are focused.

As for sensors - they definitely have non-active photosites. I saw the same 23.6x13.3 numbers, but I've also heard the sensor is 11.6MP, and 4096 4k only uses 8.85MP, and 3840 UHD only 8.3MP. So, the active area is certainly not 23.6.

Anycase, I have to do the trial and error testing by hand, and I'm getting a result of about 18mm horizontal filmBack for 3840.
 
Viddovation said:
Anycase, I have to do the trial and error testing by hand, and I'm getting a result of about 18mm horizontal filmBack for 3840.
Click to expand...

That's cool. Yes, breathing will affect the apparent focal length too (nightmare for those 3D tracking shots when focus is pulled) but that shouldn't affect you if you're basically doing motion control.

Sony were always a bit quiet about where those extra sites were. IMHO i wonder whether they're added for sensitivity rather than the actual edges of the frame.

So if you've got your camera coordinates and look at position can you not just eyeball the recorded frame with your virtual frame? Not sure what your end game is?

cheers
Paul
 
paulcurtis said:
That's cool. Yes, breathing will affect the apparent focal length too (nightmare for those 3D tracking shots when focus is pulled) but that shouldn't affect you if you're basically doing motion control.

Sony were always a bit quiet about where those extra sites were. IMHO i wonder whether they're added for sensitivity rather than the actual edges of the frame.

So if you've got your camera coordinates and look at position can you not just eyeball the recorded frame with your virtual frame? Not sure what your end game is?

cheers
Paul
Click to expand...

When I made the original post I did not have as complete a system for making adjustments after import - I was purely importing the calculated camera position - not the slider rig skeleton. The result was I couldn't easily make accurate adjustments for calibration errors. I've improved the model since.

Eyeballing it has given me the 18mm figure. But I was curious if anyone else had experience with it the same issue and had reached their own conclusion. I initially had significant difficulty getting everything to line up. In part, this was due to a Nikon Nikkor 35mm at close focus being more like 43mm! So that threw off everything else.

Endgame for the rig itself is straight forward: Be able to have camera tracking on set, and viable for shots that don't have usable reference points for tracking in post.
 
Viddovation said:
When I made the original post I did not have as complete a system for making adjustments after import - I was purely importing the calculated camera position - not the slider rig skeleton. The result was I couldn't easily make accurate adjustments for calibration errors. I've improved the model since.

Eyeballing it has given me the 18mm figure. But I was curious if anyone else had experience with it the same issue and had reached their own conclusion. I initially had significant difficulty getting everything to line up. In part, this was due to a Nikon Nikkor 35mm at close focus being more like 43mm! So that threw off everything else.

Endgame for the rig itself is straight forward: Be able to have camera tracking on set, and viable for shots that don't have usable reference points for tracking in post.
Click to expand...


Are you really only after the exact FOV of any given lens or is the sensor size itself important to you? Because if it's just FOV then you already have the tools needed to figure it out. Simply make sure your slider setup is in an easy to track environment (the opposite shooting conditions you're designing your rig for!) and do a nice move/run on it with each lens you care about. Then do a 3D camera solve in AE on the shot and it will generate the FOV of that given lens. That may be accurate enough for your needs. It doesn't take into account any lens distortion, which you should have some method of figuring out.

An approach to finding the true sensor size is to find the published pixel pitch of the sensor and simply multiply by the shooting raster size. The trick here is actually FINDING the specs for the pixel pitch. Sometimes they are incorrect and are calculated by taking the "expected" published size of the sensor and simply dividing by the raster size. Not cool.

Here's how I do things like this, when I need to accurately profile a particular sensor/lens combo for VFX work... In my case I use Syntheyes for motion tracking.

-shoot the Syntheyes provided camera distortion grid in the recommended fashion.
-Do the lens undistort process to create a perfectly rectilinear image and determine the actual optical center.
-Now shoot a very controlled moving shot (or two or three) with the lens/sensor combo. I actually have a little "stage" I made with perfectly straight and perpendicular walls and accurate measured grid lines on it. I then do a gentle, slow handheld move in the set.
-Undistort this footage and do a track. Actually, do a couple shots and check for average. Let Syntheyes determine the true FOV of the lens/sensor combo. The various test moves should agree to a good degree of accuracy.
-And finally, if you NEED the exact sensor dimensions, now that you have the true FOV and known measurements on your footage between two points you can find it... simply input the actual measured dimension between two points on your set/footage in Syntheyes and it will change the default sensor size to conform. Done!
-Wash, rinse and repeat for every lens you care about.

Regards,
 
Don't forget that the 11.9 mp sensor included beyer pattern with subsampled RGB areas. So the full sensore may be downsampled to 4k with greater color resolution.
 
You must log in or register to reply here.
Back
Top