[Bf-vfx] Footage for Tracking

[mindrones]

Hi all,

a couple of weeks ago me and Corrado Zacchi have shot some footage with
a HD CCD camera, to see if we could help providing some imagery not
affected by rolling shutter.

Enrico Gusso has been so kind to lend us a Panasonic AJ-HDX900P
(https://panasonic.ca/english/broadcast/broadcast/hd/aj_hdx900.asp) and
Gianni Centonze helped on the capture part :)

http://download.blender.org/ftp/incoming/libmv/mindrones/

Feel free to use this footage to test libmv integration as you like, and
please report any problems related to this footage, so that next time we
can shoot some better test.


We were a bit in a hurry that day so we've done mistakes for sure, but
this was a first test to hear from libmv devs. Hopefully we'll do more
later.

Some examples of mistakes to learn from:

* make sure the calibration pattern is *really* flat, taping the grid is
never enough flat (we didn't have time nor a place where to glue the
grid, sorry)

* as I've learned today chatting with Matthias: let the paper grid fill
the frame:

<MatthiasF> also, next time you shoot calibration, don't hesitate to
have the paper fill the frame
<mindrones> you mean also borders going out?
<MatthiasF> yes
<MatthiasF> we detect the inner corners of the checkerboard
<MatthiasF> so you can have those corners at the corners of your frame
<mindrones> oh crap! didn't know that!

* increase the shutter speed to reduce the motion blur (not always
possible in a real situation this is let to the director of photography
to decide this parameter)

* remember to remove any gain to avoid noisy grain (same as above)

* (to self) remember to check the iris more often, damn! :)

* also, as seen in the camera specs, the sensor effective pixels are
1280x720 and instead we've shot at 1080/25P, which means that the camera
does some image manipulation to pump the image to 1440x1080: next time,
best shooting also at native resolution to check if this helps the tracking


As an interesting test that we had no time to make, very often, even on
cine lenses, while focusing you also end up varying the field of view,
hence zooming a bit because of lenses mechanical and optical
imperfections. For this reason we didn't focus during these shots and
hence we had to choose a fixed focus distance, but this forced us to
have unfocused parts, which I'm sure won't help the tracking much.

What to do in real situations when you have to control the focus? Will
libmv take in account this factor one day?

(Not to mention real zoom of course)


Regards,
Luca

_____________________________

http://www.mindrones.com