5 or 6 tries, no results! Basic questions here

[jnz]

So... I get the idea behind photoscanning. I've used other software enough to realize I need something more powerful. I'd like to be able to scan a few car parts, sports equipment, and other non-living things.... What I've gathered so far....

1. My single Nikon D90 is fine, and I should stick to 18mm, 35mm, or 50mm but all photos should probably be taken without zooming if I used a variable lens... Correct? (Never use flash I've also understood)

2. Since I'm shooting non-moving objects, I should be fine with one camera. Multiple cameras seem to be most useful for moving or living objects. In 3D view my camera orientation seem to show just fine, so I assume I'm OK to proceed with just one camera?

3. Given the above is correct, I seem to have the option to walk around my object or use a turn table with the camera fixed.... Is one of these better than another?

4. I've tried multiple times to capture a baseball bat, kayak paddle, and other similar shaped objects. When I shoot objects, should I also have the camera pointed at the center mass of the subject? Is there ever a time I can "pan" around?

5. I've tried putting masking tape on the objects with small dots placed every so often, I've been using these to try and define points after alignment. Is this helpful or am I wasting my time?

6. Is it advantageous to correct levels and lighting in pre-processing for images? Some other software I tried specifically recommended against this.

7. Is masking something I should be looking at for either stationary or moving camera? If so, how perfect does the mask need to be? To save time I was bringing the photos in, selecting a very approximate selection, then expanding that 8 pixels to make sure everything important was included but this also includes some immediate background... Does every image need to be masked like this? It would could take quite awhile!

I think that ought to help me for now, and advice is greatly appreciated!
jnz

[tincansassoc]

Pick an object with a lot of texture and not shinny. Try putting the object on a table and walk around it taking pictures, Don't move the object. (Turntables are very challenging but possible) Masking can help but is not necessary. If your object has texture you will not need to tape on marker points. Your photos should be taken about every 15 degrees but make sure you take some higher and some lower to capture more overlapping images. You may need to capture between 30 to 80 photos, but it depends on your object.

[jnz]

Yea... I'll give it another try. I find that rowing paddle I'm trying to capture is tough because it's long and thin.

8. Can I take detail shots as long as they overlap some existing parts? For instance if I want some specific details even if those are not in the center of the part, can I pan over snap those shots then come back? Or is this something that would be better accomplished with chunks and a new main focus point on my object?

[Wishgranter]

1. Best is hard to say, wide angles make distortion - not ideal for photogrammetry...... but 18 - 50 canbe used. Speciality of pscan is that it can work with different lenses - zooms, but there are few things that should be understand......

2. Ideal is to use 2 cameras - stereo pairs, but that's not  mandatory

3. can walk, but try think set camera to safe settings - no blur and etc - low iso + high speed = good results

4. can pan around object - its know how how to do it GOOD...

5. if something like shining object you need it - from some level you cannot do it reliable - basic of photogrammetry

6. Yes, but you must understand what you doing..... sometimes its make miracles - the good ones

7. depend on objects, mostly not need, better if  work with some checkerboard under complicated objects that have not good texture - for precise alignment.

 

[Alexey Pasumansky]

Hello jnz,

1) For stable solution it is better to use the same focal length for all the photos or at least minimal number of different focal lenghts.

2) For completely static objects it's OK to use only single camera. But do not forget to provide the sufficient overlap between images.

3) That mostly depend on the task and object type.

4) It is better to point camera to the center of the object, but also you should try to use the frame space more efficiently and do not take photos from the same position.

5) There could be problems in reconstruction objects with plain untextured surfaces (especially metal or plastic). Adding irregular pattern to the surface may improve the result (for example, using color sprays, flour or something similar).

6) It is possible to correct levels of the images, however, in general cases it will not help much.

7) Masks should not be very accurate, however, you should always use masks in case of turntable capturing scenario.
Masking could be automated for fixed camera positions by using mask-from-background feature. In the other cases you can use mask-from-model feature to create mask based on rough low-level reconstruction.

You can use additional detailed shots, but in this case you need to check that focal depth is enough and no distant parts of the objects are blurred.

[jnz]

Ok, good advice everyone!

I learned of a few things I was doing wrong... I now realize that unlike other photography, I don't care about a pleasing picture more than I care about the subject being sharp and properly exposed. So, first thing I've done is to put my dslr in spot exposure mode just to make sure that if I'm angled towards the sky or towards my light source, at least the subject will still be properly exposed.

Next thing I've done is to realize that while in my head I thought I was making fairly good increments around the subject, I was really only getting about 20-30? per shot. I framed it to make sure that each 1/4 rotation of my subject was taking at least 6 images. The alignment of cameras in my latest capture looks really good and I'm doing OK at this free hand.

With no lights or studio, I've found that cloudy days are perfect for now.

I'm not sure why, but I've found that when I try and manually define markers, the subject comes out worse than if I just let PS do the work for me. Not to mention it's a TON faster to let the program do this automatically. I suspect I'm not defining markers correctly, or perhaps my photos are just getting better.

I have not yet tried to combine 2 chunks, that's next.

My latest capture of images is taking a lot longer to process.... I'm also finding that for objects, anything but Ultra High geometry is pretty much useless for me. That's too bad because it is going to take all night to render this latest section. I have another question, but I'll make a new thread for that.

[JMR]

I would suggest one more:
If your subject did not fill the image frame and most part of photo is out of interest you could add good extra-actors (simple stones may help) around your object. Filling the photo frame as much as possible you will get better results and you can mask unneeded parts after alignment is done.

[emka]

1) For stable solution it is better to use the same focal length for all the photos or at least minimal number of different focal lenghts.

Hey Alexey,

You adviced to use the min # of diff focal lenghts images. And what about diff MPx images, should the # also be min? I think of processing a scene where certain (10-15%) details should be very sharp, very high res; meanwhile, the rest and the most of the scene is needed at whatever given quality.

I have 1100D which is enough for me, and thinking of buying 550D to add details to my scene. Would that work?


And off-top: Guys (both devs and users), thank you for the software and your support on forum which at most cases cover the lack of knowledge and tutorials.

BR, Emka.

[Alexey Pasumansky]

Hello Emka,

Images with different resolution also could not be grouped together for joint calibration.