Forum

Author Topic: Single camera human captures (feedback welcome)  (Read 13581 times)

vfxman222

  • Newbie
  • *
  • Posts: 26
    • View Profile
    • Vimeo
Single camera human captures (feedback welcome)
« on: January 10, 2014, 10:48:22 PM »
Hey all, just wanted to show a result I got recently. I started experimenting with Photogrammetry last november, but have been a long time reader of the craft. Gotta say that Lee has been a big inspiration with his high quality work with full body and 4D capture work.

I work as a Visual Effects TD, compositor, 2d/3d matchmover, and ExtendScript & Python script coder. The need for more on set data over the years is what sparked the push to dive full on in with this stuff. I originally started with interior location capture using iPhone 5S, Canon Rebel XT 350D, but have progressed to Canon 5D Mark iii. These are my first ever human attempts.

Gear used:
     Canon 5D Mark iii (22mp)
     Canon 50mm f/1.4 lens

Currently working on indoor lighting setup, but these were tests and I captured them outdoors. First one during sunset, the second one during overcast day.

First attempt: 17 shots
https://vimeo.com/83652291

Second attempt: 43 shots
https://vimeo.com/83652292
--
David Torno

Lambo

  • Full Member
  • ***
  • Posts: 167
    • View Profile
Re: Single camera human captures (feedback welcome)
« Reply #1 on: January 11, 2014, 11:44:03 PM »
Wow David, this is great!. I have been trying to do the same and unfortunately the talents moved a little too much during the photoshoot and the capture didnt come even close to what you got there.
It looks amazing for only one camera!
It looks very promising. You will be able to do really good quality scans whenever you get more cameras.
Take care...
Leo

vfxman222

  • Newbie
  • *
  • Posts: 26
    • View Profile
    • Vimeo
Re: Single camera human captures (feedback welcome)
« Reply #2 on: January 13, 2014, 01:50:28 AM »
Thank you Leo. Yes, moving talent is a huge variable in a setup like this. The offset is also knowing photography fairly well and being able to have your camera setting locked in so you can move quickly around the subject to minimize the time they have to stand there. For the second experiment I did, I was able to snap the 43 shots in just under two minutes. It was also my second attempt as a co-worker messed with him and made him smile during my first pass. Regardless, this will not be a bullet proof solution. Too many variables involved to begin with. Outdoor lighting fluctuates, fluctuating light effects camera settings required to hit clean sharp images (no motion blur, deep DOF), wind/breezes mess with hair and loose clothing, etc...

I am currently getting 4k worth of lighting on Monday to help start building an indoor setup that would be more controllable and consistent. I also just got some cheap wireless triggers, so I can now incorporate a three camera setup. I am pulling in two of my older DSLR cameras (Canon 350D/8mp) as simple filler cameras. The goal this week is to attempt a full body capture with the three cameras. With this being still the R&D phase for me I am not expecting much from the 350D's, but they may be helpful, given that I was able to get passible results with a single 350D on architectural captures. Better than nothing at all I say. :)
--
David Torno

fxpop

  • Newbie
  • *
  • Posts: 4
    • View Profile
Re: Single camera human captures (feedback welcome)
« Reply #3 on: January 29, 2014, 10:24:54 AM »
Excellent results David! This gives me  hope!  :)  I was banking on using Photoscan with only 1 (or a few) camera as a body scanning system; as Kinect scanning was too low res and I didnt want to go $>25K for pro scanners. Plus, a camera  makes this portable and possible to scan anyone practically anywhere as long as they stand still.

Unfortunately, I'm not having much luck so any advice or further elaboration would be greatly appreciated.  All my results of people look something like this: http://www.agisoft.ru/forum/index.php?topic=1972.0

For that I used a Canon 5D mk II (not III if that matters) and Ilm trying to figure out what I'm doing wrong.  I think one thing I jus tlearned was that shutter speed shouldn't prevent motion blur! Before I saw that in the body guide I would have thought that would be a bad thing- so I assume Photoscan must use motion blur to compensate for tracking error between photos?

Anyway, do you know why my scans appear like that and how can I get even a facsimile of your results? Also, did you mask the images before analysis or did Photoscan use the background with  tracking? I haven't masked the images in an attempt yet, do you think that matters?

One more question, do you think an entire body scan (including fingers) is possible with a single camera?

Oh and fantastic work!!  ;D

bartosh44

  • Guest
Re: Single camera human captures (feedback welcome)
« Reply #4 on: January 30, 2014, 12:36:34 PM »
I am glad to see your topic.

I like quality of your work. It`s very good as far as one camera scan could do.

http://www.youtube.com/watch?v=3WzLsnxPiyE&hd=1
this was my try.

Did you manually modified the mesh and reimported it to Photoscan to project textures properly or is it pure scan without modification?

Have you used HDR & mask options?