Multi-camera setup project.

mphx

Re: Multi-camera setup project.

« Reply #290 on: 03 / December / 2014, 11:06:23 »

Advertisements

@H-H

jesus... you must be the life of the party

..Try human language !

« Last Edit: 03 / December / 2014, 11:32:12 by mphx »

Logged

andrew.stephens.754365

Re: Multi-camera setup project.

« Reply #291 on: 03 / December / 2014, 12:33:11 »

H-H,

yours is an interesting reference to "Automatic Disparity Control in Stereo Panoramas" and related baseline.

I found this link interesting wrt camera baseline positioning in connection with 3dme data capture www.int-arch-photogramm-remote-sens-spatial-inf-sci.net/XL-5-W1/251/2013/isprsarchives-XL-5-W1-251-2013.pdf but not necessarily all that useful.

After reading it a few times I ultimately, unhelpfully, commented, in this post http://www.agisoft.com/forum/index.php?topic=330.0 as follows:

Quote from: andy_s

Wishgranter, I still get the feeling that - rather than an optimal model - the thesis of the paper is to throw everything [including the kitchen sink] at it -

"For multiple stereo models, small such measurements can be useful for the validation of measurements with strong intersection geometry. Even though small intersection angles lead to noisy results, models with small base lines should be acquired and used within the surface reconstruction. Since large baseline models have lower image similarity - which is challenging for the matching method, small baseline models are required additionally. Furthermore, highly overlapping imagery leads to high redundancy, which is beneficial for the precision in object space."

Maybe [with a supercomputer and deeeep pockets] that is the optimal model...

Having just re-read this earlier comment from the same thread maybe someone else could find it more useful http://www.agisoft.com/forum/index.php?topic=330.msg5395#msg5395 (but i'm still not sure since it is also a bit of a fudge):

Quote from: RalfH

Quote from: Porly
The bigger the distance between stereo-pairs, then better the results. With a short basis between pictures the angel for triangulation is small what causes errors(In object direction). I was always using all pictures because I was sure that photoscan is using more Stereo-pairs to estimate one point...

Yes, this is generally true, but only to a certain extent. If the distance (viewing angle) between images becomes too large, new problems will appear, mainly related to increasing difficulties of matching features because (a) the changes are too large between images and (b) low incidence angles (i.e. grazing view of objects). In essence, image distance should be neither to small nor too large.

Edit:
but as previously linked we know, in the end, the kitchen sink is also deployed (over 200 dslr) http://chdk.setepontos.com/index.php?topic=8810.msg118503#msg118503 - thankfully not for current 3dme printing though.

« Last Edit: 03 / December / 2014, 14:35:10 by andrew.stephens.754365 »

Logged

Hardware_Hacker

Re: Multi-camera setup project.

« Reply #292 on: 03 / December / 2014, 15:18:12 »

Quote from: mphx on 03 / December / 2014, 11:06:23

@H-H
..Try human language !

The SIMPLE, and non technical, Version:-

# The !dea is to Quickly create a Thick [3D Data Exclusion] mask, for easy editing in post production.
# The Valid 3D Mesh Data is Inside the Thick Mask. (i.e Your, UN-Obstructed, Physical Work Space)
# The In-Valid 3D Mesh Data is Outside the Thick Mask. (i.e Your, Obstructed, Physical Work Space)

Thick [3D Data Exclusion] mask Is Created By:-

* The Floor Location; 3D Data.
* The Ceiling Location; 3D Data.
* The Camera Locations; 3D Data, that is already provided, by the PhotoScan Software.
* The INTERNAL Thick Mask; A Suitable Rolling Ball Size.

For The Initial Camera Rig Setup:-

* Once-Off, UN-Obstructed, Physical Work Space, Calibration.
* The 3D Data Exclusion mask, is made water tight.
* The 3D Data Exclusion mask is then re-used for all the PhotoScans.
* Re-Calibration required when the Camera Locations are changed.

For Simple Multi-Camera Rigs that have less than 360 Degree image coverage:-

* The Wall Locations; 3D Data.

H-H

Logged

mphx

Re: Multi-camera setup project.

« Reply #293 on: 03 / December / 2014, 17:04:17 »

Ok people i will sum up the situation about photoscan and masking.

For those who doesn't know photoscan...photoscan can take a number of pictures and make a 3d model out of them by finding common "points" between the photos provided.

The whole process has several stages and i will write the minutes required for each one of it in my computer.

(photoscan 1.1.0 pre-release ..improved perfomance)

1. align photos - finding points and aligning cameras ~15mins
2. build dense cloud - ~8-10mins
3. build mesh - ~3-4mins
4. build texture - ~1-2mins

Now , we have mixed two things about photoscan , masking and "projection"
That things refer to different aspects apo the whole procedure.

1.Projection

In order photoscan to find common points between photos , it needs multiple designs and colors in clothes (if we are talking about a person being shot).
He doesn't like one-colored clothes or clothes with shining material.
So if we have a person wearing such kind of clothes we have problem.Photoscan won't find points and the result would be screwed up.
So what we do ?Many different things.

1.Colored stickers all over the clothes...a bit stupid if you have a client

But it works.
2.Apply a pattern from a source (projectors let's say) ...not thick pattern so the manual removal at the end can be an easy job.
3.Apply the pattern shoot , remove it , shoot again...a bit tricky to be done if you ask me.

So ?What's the best solution?
No #1 if you shoot friends and they wont get mad with the stickers... no #2 in all other cases.

2.Masking.

Masking is the procedure where you tell photoscan to "reject" some elements of the photos from the whole procedure.
Why to do that ?Two reasons

1.Less elements in the photos , faster procedure.
2.Many times , the outline of the person being shot..the color of it ..blends with the background...and at those points the model gets ugly

So..what's the best way to mask ?
Depends on why you are masking...

IF you mask just to get rid the background for faster times...rough masking around the person is enough...

If you know that there is color blending..then a better masking is required...

Methods of masking

1.Go paranoid and mask person around pixel by pixel..done that...takes too many hours..not worth the trouble.
2.Get a shot of the person and then get an empty shot of the studio.
That requires to import every empty shot as mask to every photo. Thats a boring procedure

.Done that , but masking with this method isn't flawless , medium manual work is required.
3.Import mask from model...That means you have to do steps 1+2+3 (add times) and then choose import mask from model... it's like method no #2 ..good masking but not perfect..medium manual work is needed.
4.Do a fast 1+2 (align photos + dense cloud) and check what areas blend and need masking...Then manual mask only those areas...
5.Don't mask anything , just clear "garbages" after dense cloud.

What i prefer? Method #5 or no #4 if there is a lot of garbages and screwed up areas.

So in conclusion...masking can't help if the person wears one-colored clothes..and projection can't help with mixed colors ... two different things

« Last Edit: 04 / December / 2014, 12:19:33 by mphx »

Logged

rick

Re: Multi-camera setup project.

« Reply #294 on: 06 / December / 2014, 07:27:00 »

I made a experiment with a projector. I just used a excel sheet as the pattern.

The result was great although the surrounding light was a little weak.
Removing the “pattern points” is still a big job.

Maybe we can make a video that shows blinking points or any other patterns（such as nets ）.
So is it possbile that taking two shots one time and download last two images？

« Last Edit: 07 / December / 2014, 03:00:49 by rick »

Logged

andrew.stephens.754365

Re: Multi-camera setup project.

« Reply #295 on: 06 / December / 2014, 16:39:08 »

rick - i'll defer to a guru to, perhaps, answer directly.

Two frames to the same camera and a long time between the 2 capture events ? maybe an artec scanner (or more) and some Hao Li "non-rigid registration" (Artec studio) could be a better idea https://www.youtube.com/watch?v=7vkfyCutBjY ?

Logged

rick

Re: Multi-camera setup project.

« Reply #296 on: 07 / December / 2014, 03:15:08 »

Quote from: andrew.stephens.754365 on 06 / December / 2014, 16:39:08

rick - i'll defer to a guru to, perhaps, answer directly.

Two frames to the same camera and a long time between the 2 capture events ? maybe an artec scanner (or more) and some Hao Li "non-rigid registration" (Artec studio) could be a better idea https://www.youtube.com/watch?v=7vkfyCutBjY ?

I tried to use “nshot=2”，but there were around 5 seconds between the 2 capture events.
Yes，it was too long for a living human.

"non-rigid registration" is great.
However，it is only for Mac. http://www.hao-li.com/Hao_Li/Hao_Li_-_software.html
And，artec scanner is too expensive to me.

Logged

reyalp

13878

Re: Multi-camera setup project.

« Reply #297 on: 07 / December / 2014, 19:24:19 »

Quote from: rick on 07 / December / 2014, 03:15:08

I tried to use “nshot=2”，but there were around 5 seconds between the 2 capture events.
Yes，it was too long for a living human.

It is likely possible to get this close to the "continuous shooting" spec for your cameras. Generally this is ~1 FPS on low end cameras, so likely still too long to be practical with people.

Logged

Don't forget what the H stands for.

mphx

Re: Multi-camera setup project.

« Reply #298 on: 19 / January / 2015, 14:59:27 »

After a long time an update on the project

Things going rather well..but it was Christmas and New Years eve and people were coming to buy miniatures as presents for their loved ones.
We did 3 live interviews on 3 channels and many people saw us there and came for miniatures.
We expect work to slow down in the future..but of course we hope for the opposite.

Things out of the ordinary.

One guy asked if he could open a franchise in England...This proposition is open to discussion yet...

Another guy asked for 1:1 miniatures and we are researching the side-works that has to be done in order such a miniature can be created.

Also , today we did a test shooting with a projector projecting a pattern to the person being shot.
Results were veeeery good up to the point we are thinking to buy 3 projectors to cover all angles .
Since projectors are a bit expensive we put it on a "todo" list if things go well in the future.
For sure the result is very good and can save us some hours of manual modelling work ....

Logged

andrew.stephens.754365

Re: Multi-camera setup project.

« Reply #299 on: 20 / January / 2015, 18:08:49 »

Quote from: mphx on 19 / January / 2015, 14:59:27

We expect work to slow down in the future..but of course we hope for the opposite.

mphx, maybe it'll slow down in a year or two - but computing/hardware & you will then be set up for video and other virtual reality opportunities

Quote from: mphx on 19 / January / 2015, 14:59:27

Also , today we did a test shooting with a projector projecting a pattern to the person being shot.
Results were veeeery good up to the point we are thinking to buy 3 projectors to cover all angles .
Since projectors are a bit expensive we put it on a "todo" list if things go well in the future.
For sure the result is very good and can save us some hours of manual modelling work ....

would, sometime, like to hear test results/comment for a tailors dummy (static object) with:
(i) half camera count in continuous light&no projection (texture)
(ii) half camera count in continuous light & projection (geometry)

insofar as time saving is concerned (or lack of - in post processing).

Logged