What cameras/lenses are preferable for photoscan?

[kirk]

I tried with  8M cell phone camera and it turned into a mess mostly. Photos didn't match each other or not very good at least.    Same with cheap Canon pont-and-shot.   I tried other person photos of Nikon DSLR camera and it seems that they work better.

Does  it depend on camera or lenses at all or rather way of how pictures were taken?  Amount of noise?  Corner sharpness? Vignetting?  Aperture value?  Barrel distortion?    Maybe a specific camera brand?  I am going to buy a new camera but don't want DSLR actually.   I am looking for a kind of smaller mirrorless camera and would like to be sure it would suite photoscan tasks well.

[chadfx]

If you shoot your objects carefully and in a well lit environment, you can use lower than DLSR cameras. There might be some limits on absolute resolution and quality, but that also depends on your needs. This post has an example from a $60 Canon P&S.

http://www.agisoft.ru/forum/index.php?topic=1411.msg8361#msg8361

[ikercito]

Hey chadfx! Nice to see you around, that P&S array is going all over my head. Can we get some more info? I'm curious to see how you did it, and thinking myself of stepping in with 6 or 8 cheap Canon point and shoots. Any info on CHDK? The results you posted on the other thread are simply amazing. Of course it is a stationary object, but do you think it would work with a living person?

Kirk: For what I've learned until now (I'm a newbie), image sharpness and a high pixel count are two very important aspects, appart from noise free images (which i suppose has been the culprit in your mobile phone test) and proper lighting. One single camera is not going to cut it for moving objects, but with stationary objects you'll have better chances of getting a clean scan.

[kirk]

My cell phone has pretty wide lens and  poor  corners sharpness + vignetting.  I guess it's the main reason since the noise is the same as on all P&S with tiny sensor.  The vignetting also looks like making a kind of geometry seam from those photos which the program was able to match successfully.

 It's cool actually that the program can work just fine with cheap cameras but I still wonder what is priority factors.    They are all noisy although image sharpness may be pretty good if  lenses set not too wide.

As of multi camera setup there is a program  http://www.breezesys.com/MultiCamera/psr_index.htm
which should work with P&S. I am curious if anybody tried it.

One more thing that bother me regarding  p&S that every such camera I tried does only 4-5 shots then freezes for a minute.   Pretty annoying if you have to take a lot of shots. 
Somebody know a good  P&S  or maybe small mirror-less camera with more a less good shot-to-shot speed?

[ikercito]

Check out www.dpreview.com for camera reviews. Most of them include shot to shot times. It will all depend on your budget. Did you use a fast card in the camera? That may be the reason for the freezes. However, if you take care of focusing carefully after each photo and standing steady there should be enough time between shots not to fill the buffer, the better the camera, the faster it will write to card and empty the buffer so no freezes will occur too (unless shooting in RAW)

[kirk]

If I understand it right it's rather not shot-to-shot but mostly the buffer depth of a camera. And from what i read P&S don't use advantages of  high speed cards.   I photograph mostly  for textures , series of shots of basically  flat subjects like walls, grounds etc. to cover more of a subject  with only first shot to focus and measure and all the next just using the same values by not releasing the button completely.   
Going to use the same approach for Photoscan.   So something with better buffer would be nice.


One more thing I noticed the quality of result does not depend too much on picture resolution.    For photoscan standard version at least.  Higher resolution images takes longer to calculate but the resulting geometry doesn't differ too much.   Is it really so or it's just a matter of P&S sensors which make just more noise with more megapixels?

 

[chadfx]

Ikercito: I'm glad you found the results of the Canon P&S promising. CHDK and remote control of multiple cameras is not crazy difficult to do, but I'll see if I can prep a post of how I set it up. It runs on a lot of Canon models, so you could always pick up a cheap used one just to start experimenting and getting familiar with it (and help decide if you want to bother with going further).

I have used this setup for human facial capture and it is 'generally' fast enough and in sync enough to deliver decent results. I'll see if I can dig up a decent example (I'm using 12 P&S cameras). It's certainly not as clean as a super multicam DSLR setup, but it pretty much works. Better lighting and more cameras would help. Once again, I'm not sure if I would rely on it for high end professional use and certainly not for 'freezing' fast motion.

Kirk:

A good majority of mirrorless cameras should give you decent shot to shot times if you have a reasonably fast card. Noisy images can be larger in file size, which also can add to write times.

I also think a lot of the noise factor can be helped by just giving your subject enough light. Better lighting lets you use faster shutter speeds and smaller apertures for greater depth of field as well, while keeping your ISO set to a minimum level. All of those pretty iPhone and Nokia photos can't be fakes, can they? ;-) But they're almost all taken in good lighting conditions.

Don't underestimate the usefulness of RAW processing software like Lightroom as another tool to get more out of the images. I usually don't shy away from using LR to boost the shadows (notorious noisy geometry sources) and bring back highlight details. It also has a pretty good chromatic aberration killer in there.

Enabling the RAW shooting capability of the A3300IS with CHDK definitely allowed me to get better images off of the camera when used with LR or Adobe Camera Raw.

There is also some pretty good noise reduction software out there that can help get rid of a good chunk of the image noise and still keep enough fine detail to get a decent reconstruction. Neat Image is very well regarded:

http://www.neatimage.com/index.html


Unfortunately,  it doesn't look like the Breezesys software supports many of the newer Canon P&S cameras, so I would say that it not a viable option for a P&S multicam setup:

http://www.breezesys.com/PSRemote/features.htm#ps
The following Canon PowerShot cameras do not support remote capture operation and therefore will not work with PSRemote:
A series: A410, A420, A430, A450, A460, A470, A480, A490, A495, A530, A540, A550, A560, A570 IS, A580, A590 IS, A610, A630, A650 IS, A700, A710 IS, A720 IS, A800, A810, A1000 IS, A1100 IS, A1200, A1300, A2000 IS, A2100 IS, A2200, A2300, A2400 IS, A3000 IS, A3100 IS, A3300 IS, A3400 IS, A3400 IS, A4000 IS
Digital ELPH/IXUS: SD30, SD40, SD400 (Digital IXUS 50), SD430, SD450 (Digital IXUS 55), SD500 (Digital IXUS 700), SD550 (Digital IXUS 750), SD600, SD630, SD700 IS, SD750, SD770 IS, SD780 IS, SD790 IS, SD800 IS, SD850 IS, SD870 IS, SD880 IS, SD890 IS, SD900, SD940 IS, SD950 IS, SD960 IS, SD970 IS, SD980 IS, SD990 IS, SD1000, SD1100 IS, SD1200 IS, SD1300 IS, SD1400 IS, SD3500 IS, SD4000 IS, SD4500 IS
SX series: SX500 IS, SX260 HS, SX230 HS, SX210 IS, SX200 IS, SX160 IS, SX150 IS, SX130 IS, SX120 IS, SX50 IS, SX40 HS, SX30 IS, SX20 IS, SX10 IS, SX1 IS,
G Series: G11, G12, G15, G1X
S Series: S90, S95, S100, S110
Other: TX1, E1, D20, D10, D20


And finally, there is definitely a noticeable difference in build quality, even with these small sensor P&S cameras. This image has the ultra high build on the left and the high quality build on the right. Depending on your needs, this difference might not matter, but you can squeeze a bit more out of these little cameras if you really want to. It's just a matter of taking your time to shoot well.

Anyway, just a few more thoughts to ponder for you...hope it is helpful!

Cheers, -Chad

[kirk]

Thank you for a lot of info chadfx,    Looks like that PSromote is a kind of abandon-ware and is not going to go further. I curious if there is anything similar for newer cameras? 

[chadfx]

I'm not sure if there is a polished 3rd party product that can do it all. A few of us around here have explored other forums like the CHDK ones:

http://chdk.setepontos.com/index.php?action=forum

and there certainly seem to be some people writing code and specialized hardware to work with P&S cameras.

Some people have also used PTP functionality to control their cameras, but once again, I'm not sure if there is a tool out there for solid multiple camera support:

http://chdk.wikia.com/wiki/PTP_Extension

It seems like most of the major camera remote software is really set up for Canon & Nikon DSLRs only, unfortunately.

[chadfx]

These guys also build some specialized gear to control a variety of cameras and other equipment, but I don't know if anyone is using these on a large scale:

http://www.gentles.ltd.uk/products.htm

[chadfx]

As promised...this was from a test with 12 of the Canon A3300IS cameras. The setup was not exactly ideal, as a couple of the cameras were accidentally set a bit wider in focal length than the others.

The ultra high res build is on the left (with the default mesh interpolation that has left a couple of holes where the data was too incomplete) and the high res build on the right (with the extrapolation setting filling in those holes). The ultra res definitely shows more noise issues than the doll example I posted earlier, but it does also show more fine detail (plus I hadn't shaved recently ;-)).

Probably better lighting in the photos would yield a more noise free result. I'd still like to tweak the camera positions to find a good arrangement that will create a solid model without needing to extrapolate past the hole issues.

[ikercito]

That looks great Chadfx! I really think that an investment in a few point and shoots can yield very good results... At least for the level I'd need to. There is some noise and a few holes in the high res, but a good Zbrush cleanup can probably fix it. Does CHDK allow sync shots with all the cameras supported, or just specific ones?

Really looking forward that post with instructions on how to setup this kind of arrangement. Congrats!

[kirk]

It seems to me that the last 1.0.0 build of photoscan standard makes more noise then some older one I have used before.   Do somebody experience the same?
 

[MichaelW]

We have found that we are getting the best results when utilizing a wide angle lens DSLR setup (Canon). We are working in cultural heritage documentation so we are dealing with larger scale sites. It all depends on what you are trying to do. I've had success with a stock 18-55 Canon lens that came with my SL1 and I've had success with my Powershot G10.

[kirk]

How much wide?  Does it cope well with usual barrel distortion of wide lenses?