Icomprehensible bad stitching

[Josep Brunet Insensé]

Hi,

I have observed some inexplicable stitching errors that I have not been able to solve or entering control points manually.

Sample 1:

Sample 2 (equirectangular panorama):

There is no parallel error. The nodal point is perfectly calibrated.

The checkpoint optimizer indicates "very good".

12 HDR images taken with a tripod

Sigma 8 mm fisheye lens,

Nodal Ninja R10

PTGui Pro 11.13

Link to project, panorama and image files (Google Drive): https://drive.google.com/drive/folders/1uLNpIyuG1TXUneneDCI0_NZsExTX4U-t?usp=sharing

What can be the cause?

Is there any solution?

Thank you very much in advance,

Josep

[John Houghton]

On Wednesday, July 31, 2019 at 6:17:44 PM UTC+1, Josep Brunet Insensé wrote:

Hi,

I have observed some inexplicable stitching errors that I have not been able to solve or entering control points manually.

Sample 1:

Sample 2 (equirectangular panorama):

There is no parallel error. The nodal point is perfectly calibrated.

The checkpoint optimizer indicates "very good".

Your control points are very much clustered along the centre band of the panorama, with relatively few on the ceiling where the misalignments errors show. However, it's not possible to eliminate all parallax in the case of the fisheye lens owing to the entrance pupil not being located at a fixed point. (It's the entrance pupil about which you need to rotate the camera, not either of the two nodal points).  The apparent position of the entrance pupil varies for light rays entering the lens at different angles to the lens optical axis.  Even at the horizontal level, there is parallax visible, though this would be more evident if there was something much closer to the camera. See the animated gif that can be downloaded from https://we.tl/t-SKRUH9vAfJ that shows this clearly. The setup of the panohead is necessarily a compromise.  With care, you may be able to get a good enough alignment at the zenith as well as at the horizontal level.  Try adjusting the camera position a mm or two on either side of its present position to see if an improvement can be obtained.

John

[Josep Brunet Insensé]

Hi John,

Thank you very much for your explanations. There are some complicated things that I have not understood, but the basics (maybe) yes.

I tried to manually enter additional control points in PTGui on the ceiling where the misalignments errors show. But I couldn't solve it in any way.

Do you mean my main problem is here in the point of this rail?

I tried to calibrate using this procedure:

(https://youtu.be/AkCxhlXKFMo)

With a pencil on a pole 50 cm in front of the lens and using a vertical line 3 meters further to achieve a total match. The results seemed good (no parallax) rotating the camera (to the point that allowed me to view live through the LCD with zoom). But maybe I thought it was fine and it wasn't.

Now I will try to calibrate again using this procedure, with two poles:

(https://youtu.be/IFQHoCjFTn8)

Is this what you mean when you write to me "Try adjusting the camera position a mm or two on either side of its present position to see if an improvement can be obtained"?

On the other hand, do you mean that with a fisheye and a ringhead and 4 shots (12, to make an HDR) I will never get totally satisfactory results?

Thank you very much for your help,

Josep

El dimecres, 31 juliol de 2019 22:59:43 UTC+2, John Houghton va escriure:

[Trevor Curtis]

Hi Josep

As you rightly say there are complicated things going on inside of fisheye lenses but take a look at this article ......it may shed some light on the principles associated with determining better no parallax or limiting the issue

http://michel.thoby.free.fr/Fisheye_history_short/Beyond-the-pupil.html

you will also see a link to Johns more detailed explanation of Parallax about half way into the article too

hope this helps with the theory and gives some explanation as to a "compromise" solution

Trevor

[Erik Krause]

Am 01.08.2019 um 12:10 schrieb Trevor Curtis:
> As you rightly say there are complicated things going on inside of fisheye
> lenses but take a look at this article ......it may shed some light on the
> principles associated with determining better no parallax or limiting the
> issue
>
> http://michel.thoby.free.fr/Fisheye_history_short/Beyond-the-pupil.html

Or even better here:
http://www.pierretoscani.com/fisheyes-(in-english).html

Half way down the page there is a nice animation...

--
Erik Krause

[John Houghton]

On Thursday, August 1, 2019 at 10:10:07 AM UTC+1, Josep Brunet Insensé wrote:

Hi John,

With a pencil on a pole 50 cm in front of the lens and using a vertical line 3 meters further to achieve a total match. The results seemed good (no parallax) rotating the camera (to the point that allowed me to view live through the LCD with zoom). But maybe I thought it was fine and it wasn't.

Now I will try to calibrate again using this procedure, with two poles:

Josep, Your method should yield an acceptable result, yet instead of the second more distant pole I would recommend using a feature as far away as possible e.g. something in the far distance seen through your window.  The reason is simple: BOTH of your indoor poles will exhibit parallax shift as you rotate the camera, so what you observe will be the difference between the two shifts. Something in the far distance will suffer negligible shift so you will then effectively observe maximum shift on the near pole.

John

[Josep Brunet Insensé]

Thanks Trevor, Erik and again John,

All your explanations are very interesting. I was unaware of these variations that occur in fisheye lenses. All this will involve new studies on my part, new tests and even studying if I need to invest in new equipment to offer professional results (if this is impossible with my current equipment). Maybe new pano heads other than the ring that allow me to take four more shots at 45 degrees up and 45 degrees down. According to what I have read (thank you), the fact of adding those eight planes up and down will offer me more reliable results although for each photosphere I will have to take 36 raw photos (to make the HDRs).

And maybe even the possibility of using other lenses (which I already have), such as a 16-35 mm Canon in full frame cameras that offer me a better quality and a better resolution although all that will also increase the time of my workflows.

What is clear is that I cannot work with the insecurity of offering my clients results on the ceilings like the ones I showed here.

Indeed, John I am going to do the calibrations that you mentioned taking as reference the page that Trevor has provided me (http://michel.thoby.free.fr/Fisheye_history_short/Beyond-the-pupil.html) and specifically the one used for Calibration a reference to a farther point (http://www.johnhpanos.com/epcalib.htm).

(screenshot)

I will return here to comment the results.

Thank you all for your support.

Josep

[Josep Brunet Insensé]

Screenshot (sorry)

[John Houghton]

On Thursday, August 1, 2019 at 4:04:09 PM UTC+1, Josep Brunet Insensé wrote:

And maybe even the possibility of using other lenses (which I already have), such as a 16-35 mm Canon in full frame cameras that offer me a better quality and a better resolution although all that will also increase the time of my workflows

Josep, Yes, the 8mm lens is not an ideal choice for a fullframe camera since half of the pixels are wasted on black space.  The Canon 8-15mm fisheye is excellent, as it permits the useful 12mm setting as well as 15mm.  Also take care to remove chromatic aberration in raw processing, which makes such a difference, as shown in the attached crop from near the edge of the image circle.

John

[Josep Brunet Insensé]

Hi John,

Thank you very much for your advice. I know the Canon 8-15 mm. In any case, if I consider buying a panohead of the Nodal Ninja NN6 type for example, then I can use many other lenses that I already have.

I have to study all this.

I will start by trying to fully calibrate my panohead r10 with the fisheye. And then go step by step collecting and studying all the information received.

I will write here all my progress.

Thank you very much again !!.

Regards,

Josep

El dijous, 1 agost de 2019 17:46:12 UTC+2, John Houghton va escriure: