New User Problems with Body scanning

[fxpop]

Hi,  I just bought Photoscan and love it!       But I'm running into problems as I'm starting to learn it, the Photoscan noob that I am.

I've had success with scanning basic inanimate objects like small statues:
http://imgur.com/a/IiHaJ#0.jpg

(There is some bumpiness to the scan, but it is totally usable and good enough for me to bring into another app like 3DCoat and retopologize it and texture bake it smaller if I wanted - though I suspect the bumpiness I am experiencing is user error.)

My problems arise whenever I try to scan a person.  No matter what I do I keep getting a very bumpy surface that makes the models unusable.

I'm not sure if this is a calibration issue, a camera quality issue, a color tracking issue with noise from the images, or me having the settings wrong when aligning, building the point clouds and mesh.  I'm assuming the bumpiness is not because of micro movements in the person breathing or whatever, because there is not bigger bumpiness in the chest area where one would expect the most moving.

I have tried scanning people with a 12Mp Sanyo HD2000 camera as well as a Canon 5D mkII and both have similar issues so I am expecting the problem to be something I am doing wrong.

I am scanning with 1 camera, basic indoor lighting, and also lighting using one of these:
http://www.amazon.com/NEEWER%C2%AE-Dimmable-Digital-Camcorder-Panasonic/dp/B004TJ6JH6/ref=sr_1_1?ie=UTF8&qid=1390792612&sr=8-1&keywords=cn-160+led
while taking pictures handheld - but I am careful about focus and overlap between pictures.

I'm talking particularly about the bumpiness on the skin and face and arms...

Here is an example of how it turns out:


I would greatly appreciate any help in workflow or settings to get a more acceptable output.  Obviously, I'm not going to get the results that the guy who uses 80 DSLRs on tripods in perfect studio lighting, synced together with 1/10000th sec shutter, but surely a handheld series of photos can do better than what I've done.  Perhaps if someone could help me with calibration that would be useful?  I tried using the lens app from here, but kept getting errors trying to import its xml into Photoscan - so I tried importing the checkerboards into Photoscan itself as a different chunk but couldn't figure our how to get it to use it all correctly.

Thanks very much for your help!!

[FoodMan]

you're lucky to have that result so far... but I think that's about how far you can go with a one Cam setup..

so to answer, unless you photograph a dead person, he or she will always move... always, no matter how stand still a person might be good at..

.. yes you do need a multi cam setup... 

f/

[fxpop]

you're lucky to have that result so far... but I think that's about how far you can go with a one Cam setup..

so to answer, unless you photograph a dead person, he or she will always move... always, no matter how stand still a person might be good at..

.. yes you do need a multi cam setup... 

f/

Thanks very much for your reply.

But I'm confused... wouldn't the "micro-movements" that the person is moving with be particularly in the chest area (where breathing is)?  I would expect that area to have by far the most movement and thus more displacement in the noise around that area?  The entire person seems equally noisy - so I don't understand where this noise is coming from.

Does the displacement noise come from movement of the person? jpg noise in the image? blurriness of photo? (I did not include images to analyze where the image was not perfectly sharp)

What did you think of the statue image above?  That had bumpiness, but not as much (also a smaller 12Mp camera)  How could I improve that one? If it was the movement of scanning a living being then I would expect my statue scan to be a little cleaner?

Sorry for all the questions, I am trying to learn what is possible.  Feel free to link to another post if I have missed something and it is already explained here.

I would love a multi-cam setup, but for now I want to see what is the best I can with 1 or 2 cameras not synced...    I don't mind a little clean up in 3DCoat or something.

thanks for any help!

[Alexey Pasumansky]

Hello fxpop,

I think the following topics related to the head/body scanning with only a few cameras may be interesting for you:
http://www.agisoft.ru/forum/index.php?topic=1896
http://www.agisoft.ru/forum/index.php?topic=970

I also recommend to check Full Body Capture tips, if you haven't already seen them: http://downloads.agisoft.ru/pdf/Image%20Capture%20Tips%20-%20Full%20Body%20Capture.pdf
There are useful recommendations that could be applied even for non-living objects like statue (for example, more effective use of image frame space, better lighting, minimal ISO and etc.)

For your tests of body capture with the single camera I can recommend to ask the person to sit next to the wall, it will help to minimize head shakes during the capturing session, because breathing also causes head and shoulder movements.

As for the camera calibration, I don't think it will improve the results. But it's strange that you have problems with the calibration import. Probably, you are using Import Cameras option in Tools menu instead of Load button in Camera Calibration dialog.

[bartosh44]

Hi,
I am also trying to get best result with one camera.

I`ve tried twice.

First: http://www.youtube.com/watch?v=gsz_bbm88_Q
Second: http://www.youtube.com/watch?v=3WzLsnxPiyE

I hope third will be better

I used Zbrush to play with the mesh and I reimport mesh into PhotoScan after UV to help scanner wrap textures on existing mesh and omit freeze during building atlas. High density details you can project from textures if you have good wrap on model and good uv. I hope it helps.

Edit: From one camera scan geo is rough. You have to smooth it, decimate it, make uv on it and reimport to scanner. Then you will have overall shape and good texture wrapped on it. You can`t scan pores from one camera to geometry in my opinion becouse of micro movements of human body, so best method for me is use wrapped texture and change it to high details

[Steve003]

Hi Fxpop,
I am getting exactly same as you are, material a bit bumpy, t shirt with logo and text readable, skin though (slight hairy arm greckles, ideal varied colouring etc) VERY bumpy and useless, a total mess, and thats even for the upper arm that I am sure did not move, which is a bit better than the forearm.
Canon R6 Mk II 24mm f8 1/500 iso 5000 in full shade, 7pm

Have you found the problem ?

There was no breeze that evening.

3 secs per shot, even taking a sequence of 20 pics of the top and outer area arm sees a mess occur, even editing out the frrass and bushes before make model.

I have planned a massive few thousand photos photoshoot and this is critical and its doing this to me.

Hired all this gear and I am getting these results same as you.

I wonder if photographing around the upper arm first then some obliques, and then photo the floor as a 'end of sequence marker' then photo around the forearm, and some obliqies, then photo the floor again, then photo the hand, same method, and the floor, three projects, then bring them together in blender, and knit them together.
Dont attempt to get a mesh and texture as its impossible die to micromovements to do so.
Agisoft is unforgiving on any microvements I would say, Alexey correct me if I am wrong !

Has anyone run the same photos through Photoscan then Metashape, are the results the same ?

also mirrorless camera for low light , 400 shots battery dead, have 17000 to do. usless cameras, Dslr 2000 on one battery.

Steve003