Problem panorama like StreetView

[Anna Jagura-Szczyglak]

I use a Ladybug 6 camera system to capture 360° panoramas. However, I would like to supplement each panorama with an image of the area directly underneath the vehicle, similar to the effect used in Google Street View.

The images are captured every 5 metres and have precise georeferencing data. My goal is to use the forward-facing image from the previous capture point together with the two rear-facing images from the next capture point to reconstruct the area hidden beneath the vehicle.

The problem is that when I add and calibrate control points between the previous and next images, the entire panorama becomes misaligned and distorted.

I have attached the project file together with the complete image set. Could someone please advise me on the correct workflow, explain what I am doing incorrectly, and suggest how this issue can be resolved?

I would also be willing to pay someone to prepare a correctly configured project file for me.

[Anna Jagura-Szczyglak]

[Anna Jagura-Szczyglak]

[Anna Jagura-Szczyglak]

[Roger Oberholtzer]

We are doing something similar:

We have systems on vehicles with two ladybugs (5,5+,6), one in the
front of the vehicle and the other on the back. It's on a Sprinter
van, so keeping the vehicle and such things out of the image is not an
easy task. They are synchronized to take images when each camera is at
the same location, using a wheel pulse encoder to track distance and
thus when each should trigger. Of course this is not a perfect system
as there can be parallax issues. But most often the join between the
images is not seen. We use a high grade GNSS/IMU receiver that lets us
post-process locations and orientations using virtual base stations no
further than 20 km from where the image was collected. It's the IMU
data we need. This is because vehicle orientation will change between
the two image captures. We then combine the two images (rotating as
needed) into a single image that has very little of the vehicle in it.
The vehicle has lots of other equipment that would otherwise fill the
image.

For all this to work, at the core level, we rely on a wheel pulse
transducer to control when each image is taken. If the images are not
taken at the same location (parallax aside), combining the images will
be problematic.

BTW, we looked at ptgui to help with all this, but found that our own
solution worked better in our scenario.

> --
> You received this message because you are subscribed to the Google Groups "PTGui Support" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to ptgui+un...@googlegroups.com.
> To view this discussion visit https://groups.google.com/d/msgid/ptgui/dcf68c45-c782-49c0-92b7-8bb7593b1ba9n%40googlegroups.com.



--
Roger Oberholtzer

[John Houghton]

Anna,  The mistake you are making is you are not keeping to the rule that when you are adding control points to a viewpoint image, you must place them only on a flat area (like the road).  You have created points on the road and on trees etc.  My attempt (project file attached) was moderately successful:

John

[Anna Jagura-Szczyglak]

Thank you very much for your help and for pointing this out. I now understand that the control points should only be placed on flat surfaces, such as the road, and not on trees or other objects.

Would it be possible to fill in the missing areas in the image by adding another 2-3 photos? If so, would you be able to check whether this would work and, if possible, help me do it?

I would really appreciate your help with this.

Best regards,
Anna

[Anna Jagura-Szczyglak]

[Anna Jagura-Szczyglak]

poniedziałek, 10 sierpnia 2026 o 18:29:02 UTC+2 Anna Jagura-Szczyglak napisał(a):

[John Houghton]

On Monday, August 10, 2026 at 5:28:29 PM UTC+1 Anna Jagura-Szczyglak wrote:

Thank you very much for your help and for pointing this out. I now understand that the control points should only be placed on flat surfaces, such as the road, and not on trees or other objects.

Would it be possible to fill in the missing areas in the image by adding another 2-3 photos? If so, would you be able to check whether this would work and, if possible, help me do it?

Anna, I made an attempt to add in your extra 3 images, but found it difficult to get a satisfactory result.  The image quality is not good enough to identify matching clear features in the road surface to get a good spread of control points for viewpoint purposes.  It seems odd to have such a variation of lens types for the various images, and there is no exif data to confirm the settings in your supplied project file.  The attached project file derived from that is as far as I got, and though it shows a more complete nadir area, there are clearly problems remaining.  I'm afraid this doesn't seem to be a practical solution for your requirements as put forward in your original post.

John

[Philip Chong]

If you get it close enough, you can finish it off with some AI of fish and chips.  

Or maybe thanks but no thanks. hahaha