Region Orientation, no fine control..

[Infinite]

Would it be possible to get finer control over the region box orientation?

As it's our only method to have control over scene orientation and position using a script, we really need finer incremental control over it. At the moment when adjusting the rotation, the increments are just too large and it's VERY hard to do fine adjustments when using the region box to position and rotate a scene using "Coordinates to bounding box + rotate" script and the region box..



The same goes for the viewport, the rotational increments are very wild and too large for fine control.

Would this be possible?

Lee

[JMR]

++++1

[James]

Infinite, have you tried the coded targets? They are great, and easy to print off from inside PS.

Place 3 (minimum) or 4 to the floor around your subject, maybe in a 1m square, measure them in and stick them down firmly somehow.

Automatic marker detection will whip though and identify the markers in each image.

You then manually enter the coordinates (X,Y) as measured, Z = 0, click update and the model is automatically oriented and scaled. You can then export the markers coordinates and reuse for future projects so no more manual entry.

This would also give a more accurate scaling than your current method since scaling a 2m object by a 1cm reference will result in tiny inaccuracies being multiplied up significantly (200x).

Using the markers has no effect on camera alignment so long as you don't try to optimise using them, and i wouldn't recommend that in your case (unless you have markers high up as well as on the ground) so you have nothing to lose!

This whole thing could probably be incorporated into a python script too.

[James]

But I do agree that sometimes you just want to have fine control over rotation of the model and/or bounding box.

I use the Leica Cyclone point cloud processing software which gives fine rotation control while holding the ALT Key so that's one way to do it.

[Marcel]

Another easy solution we would be a size setting for the trackball in the Preference Window. If the trackball was two times as big, it would be more precise.

(For rotation perpendicular to the camera you can already grab anywhere in the screen, so if you grab at the edge of the screen it's more precise. But that is not really an option if you need to lots of rotations along a random axis).

[Infinite]

Infinite, have you tried the coded targets? They are great, and easy to print off from inside PS.

Place 3 (minimum) or 4 to the floor around your subject, maybe in a 1m square, measure them in and stick them down firmly somehow.

Automatic marker detection will whip though and identify the markers in each image.

You then manually enter the coordinates (X,Y) as measured, Z = 0, click update and the model is automatically oriented and scaled. You can then export the markers coordinates and reuse for future projects so no more manual entry.

Using the markers has no effect on camera alignment so long as you don't try to optimise using them, and i wouldn't recommend that in your case (unless you have markers high up as well as on the ground) so you have nothing to lose!

This whole thing could probably be incorporated into a python script too.

Thanks for the reply. Do you have any examples of this process? A video tutorial showing the whole process could really benefit the community, it sure sounds interesting. Although I'm not sure I quite understand. Things that sound good in theory more often than not don't work well in practice.

You only ever need to do the scaling and alignment once per capture session (unless you move your cameras), as you can re-use the camera alignments.

This would also give a more accurate scaling than your current method since scaling a 2m object by a 1cm reference will result in tiny inaccuracies being multiplied up significantly (200x).

What do you mean scaling a 2m object? FYI it's not my process but the one outlined by Agisoft.

[James]

I shall try coming up with a tutorial, that's a good idea.

What i meant by the scaling issue is that your object is generally a person, so ~2m (or near enough).

If you misplace your markers on a 1cm reference object by, for example, 0.5mm then photoscan will be scaling each 10.5 or 9.5mm in you scene to 10mm, or a 5% error, which over the height of a person could be as much as 100mm, or 4" off their height.

With markers further apart, then if your reference distance(s) is 1m then 0.5mm inaccuracy is only a 0.05% error, or 1mm over the height of a very tall 2m person.

[James]

I have created a PDF tutorial Coded targets for alignment and scale on dropbox here: https://db.tt/ZZOFJ59u

I only just spotted the agisoft one which deals with coded targets for scale only here: http://downloads.agisoft.ru/pdf/PS_1.0.0_Tutorial%20%28IL%29%20-%20Coded%20Targes%20and%20Scale%20Bars.pdf

Had I noticed that first I might have made more attempt to match the format.

I also uploaded a zip containing everything including project file and photos here: https://db.tt/jiorP0ii

I apologise that the subject matter is far from inspiring! If there is any demand I shall redo it with something more interesting, this was just the first thing I found and wanted to do it while I thought about it and still had the time!

[Infinite]

This is very generous of you and quite an insight, thank you James. REALLY useful. I will test this method out and report back in the coming weeks. My studio is in bits at the moment. Upgrading.

Big thanks 

[Andrew]

I second Lee's feedback about region rotation, it would be nice to have more control there!

As for the method of using coded targets to align scene, kudos to you James!

Perhaps Agisoft can implement similiar workflow, doing the math automatically? Perhaps it could be as simple as defining that first two markers represent X axis and line from third marker, perpendicular to the line connecting first two, is our y axis?

Maybe we could even place markers at even distance from the scene center (around scanned subject) And have photoscan treat midpoint between markers as the 0, 0, 0 origin?

Lastly, I really wish that manual markers, rulers and coded targets features would trickle down to Standard Edition, leaving all of the advanced ground control, ortophoto and other DEM-related features, plus 4D, to differentiate Proffesional Edition...

[James]

Glad that it is proving useful

A even easier way without requiring any maths is just to set up the 3 markers in a right angle triangle, so either line them up in autocad or photoshop etc and print them out using a large plotter, or print off individual markers and align them to something you know is 90 degrees, i.e. large piece of paper, carpet tiles, cardboard box, or do something clever etc. Then you only need two measurements, one between origin marker and X marker, and one between origin marker and Y marker.

[maddin]

I also uploaded a zip containing everything including project file and photos here: https://db.tt/ZZOFJ59u

That last link points to the PDF, not the ZIP file...

M.

[James]

Well spotted! I have now corrected it above and here it is too:

https://db.tt/jiorP0ii

[maddin]

Well spotted! I have now corrected it above and here it is too:

https://db.tt/jiorP0ii

Great, thanks a lot!

Martin