Colour Color Scans Using External Camera 360 Theta Z1

[D2RSURVEY]

Hi

I use a ZF 5010, doubtless the M-cam was ahead of its time at inception but by the time I acquired it, well we will say no more. I did rig a DSLR with fish eye lens on a nodal ninja but for this one needs some photography skills and patience for the pano stitching which isn't 100%. Up steps the Theta Z1 with Dual Fish Eye app.

It took a while to rig it to the correct height, Ricoh provided the exact height, I have that somewhere though I'd recommend going for the Theta X with hot-swap battery for ease of use with minor quality compromise. Using the Dual Fish Eye app on a Nokia I can go around the corner and activate the shutter. Using the smallest aperture and 9 HDR brackets I have illuminated the dingiest of corners. First I tried a 90deg turn and took another shjot to combine all in PTgui however Lightroom with the Theta Stitcher plugin is the way forward.

So one press photo on site, develop in Lightroom, edit with Theta Stitcher pluging, export correctly sized pano into the ZF project directory. You can get through quite a few while Laser Control is on with a plane to plane.

IMG_4776(1).jpg

Problem

Though the panorama to scan correlation process completes without hitch, it has a slight twitch. So here the red cone is projected on the blue barrel in the background. The accumulation of these error makes for a messy colour unusable pointcloud. The walk around in Autodesk Recap is still fine + I draw in intensity but it would be good to produce a decent colour ortho now and then.

Cone.png

Question?

Do I need to produce an Intrinsic camera calibration file? I find that this routine runs for a while and then stops having produced nothing. I cannot work out what the software wants, there are only so many ways to photograph a checker board.

Intrinsic.jpg

There is the pass pair definition, AKA "error 8 Not enough corresponding points" even though you have 7 targets matched scan to pano. I do generally manage to figure this one out but there has to be an easier way. Its almost right.

[D2RSURVEY]

OK having another go, I opened a project and copied the calibration images files into the project directory just in case. I can see that the software rotated the images as requested but beyond that we are stuck. I didn't expect this set two work as they are the dual lens images. Note you need the single lens app to shoot from one side only. Hmm ram and CPU are still quivering, do you leave it, is it crunching or has it crashed?

Rotate.jpg

[PatHughes]

Hi Wayne,

Thanks for posting your progression on this subject, I'm very interested to see if you can get the Zeta calibrated and colourising the point cloud accurately.

With the new layout of LSF I'm not sure if the Z+F team will pick this thread up straight away.

Send an email to [email protected] and they will give you some guidance on how to setup the camera properly in LaserControl.

Please come back here though and let us know the results, I am sure there are other 5010 users who could benefit.

[D2RSURVEY]

Hi I shouted out to the UK tech guy but got an email bounce, I'll enquire again on Monday.

Firstly my checker board photos are rubbish so I have started over. We need around 20No of them at varying angles to the board which should mostly fill the field of view. Reject any photos where the board is partially missing.

As Laser Control offers no feedback loop I have been testing the photos in other software. My thinking is if they are acceptable to other software then they should be to laser Control.

This company has a calibrator and checker patterns to download calib.io

Up to now that software fails to detect their own checker board pattern! Shout out to their tech support.

Moving onto to a Matlab trial, desktop version with Computer Vision Tool Box installed. This has a Camera Calibrator App.

https://www.youtube.com/watch?v=vhi7rbYWShk

Matlab SS.jpg

OK so as top right my images are still not good enough and I need to get the Mean Error in Pixels below 1 or at least the you tube guy managed. So then knowing I have a good set, plan A is to represent this set of images to Laser Control knowing they are good to go for it to generate its own calibration file. If the software still decides to do nothing.

Plan B Export Fisheye parameters from Matlab, Export a calibration XML from Laser Control, edit the XML with the Matlab data and re-import to Laser Control. Bingo no more number plates on the tarmac, maybe.

[D2RSURVEY]

Matlab SSa.jpg

OK down to below 2 pixels, intrinsics as above, algebra below. Laser Control didn't play ball with the same images.

Intrinsics Explaination.jpg

Right I'll have to leave that there for now, my server has shut down. That means its definitely past bed time.

[D2RSURVEY]

Mablab camera intrinsics as below with algebraic variable order.


fisheyeIntrinsics with properties:

MappingCoefficients: [762.9848 -7.8773e-04 1.4443e-06 -1.3069e-09]
ImageSize: [1920 1920]
DistortionCenter: [980.9257 909.7651]
StretchMatrix: [2×2 double]

Properties
expand all
MappingCoefficients — Polynomial coefficients for projection function
[a0 a2 a3 a4] vector

DistortionCenter — Center of distortion in pixels
[cx cy] vector

Unfortunately having copied an existing .ini calibration file and changing it to a spherical camera, then exporting the XML as below there is nowhere to input the intrinsic information. This could be the end of the line for my skills. I suppose I could export the individual scan external calibration XML and tweak the roll, pitch, and yaw which is where the error is. Re-import and apply, yeah that could take a while.

Note: XML file opened in Notepad++, I have tested editing the camera models and re-importing to laser control, this works fine.

<?xml version="1.0"?>
<Cameras>
<Camera name=" Theta" manuf="Ricoh" typ="Spherical" no="" serialno="" revision="1.0">
<model>Z1</model>
<optMode>-1</optMode>
<date>2018 24 02 - 00:34</date>
<info>Size 10000 x 5000</info>
<sphExt_w>5000.000000</sphExt_w>
<sphExt_h>2500.000000</sphExt_h>
<sphExt_off_u>0.000000</sphExt_off_u>
<sphExt_off_v>0.000000</sphExt_off_v>
<sphExt_off_z>0.000000</sphExt_off_z>
<sphExt_useAutoCalibVR>0</sphExt_useAutoCalibVR>
<R0_roll>0.000000</R0_roll>
<R0_pitch>0.000000</R0_pitch>
<R0_yaw>0.000000</R0_yaw>
<R0_tx>0.000000</R0_tx>
<R0_ty>0.000000</R0_ty>
<R0_tz>0.000000</R0_tz>
</Camera>
</Cameras>

To rule out one more possible error, my set up if anyone or ZF has time to run the create new calibration attached are the Matlab image set.

Thanks

[D2RSURVEY]

Hmmm am I barking up the wrong tree? Do I need a lens intrinsic calibration file or has that job already been done in that I am presenting ZF Laser Control with a Equirectangular panorama of the appropriate pixel W x H. 10000x5000 for hjigh res and 5000x2500 for medium res. The intrinsics were needed to stitch the dual fish eye images which was done in the Theta stitcher app and Ricoh should know their own camera calibration.

This image highlights the problem as the above cone but without the multi scan noise. This is in Recap, as you can see it just needs a tweak down and to the right. Or is it that there is fault in the pano I am presenting or that Laser Control is miss matching the features? I have found some settings in Laser Control however they don't appear to make any difference.

Scan Image Offset.jpg

Default Colour Settings.jpg

[D2RSURVEY]

Trying Passpair Definition

PassPair.jpg

Well its moved but it is still not right

After Pass Pair.jpg

[D2RSURVEY]

OK it turns out that not all panos are stitched the same. I am getting closer with this 2x2 dual fish eye, developed in Lightroom but stitched in PTGui. I could do with an example of an automatically stitched scan. I'll have to have a look way back at an M-Cam example.

2x2 PtGui.png

2x2 PtGui.png