Content missing after stitched for thermal images taken by a drone

[Alex]

Hello, 

Using a drone, I have taken a series of images named: frame_0001.jpg to frame_0010.jpg (as per the attachment). 

  Noted: the meta data of each file has been removed. 

Then using Hugin, I have created the panorama image "panorama_by_hugin.png".

But the result is not satisfactory, because I have found-out that: out of the original 15 solar planes, only 8 are captured in the final panorama image. 

I expected I can achieve to create a panorama image like "expected_panorama.png".

Can someone help me with the following queries: 

1. Is the missing of certain content in the final panorama image normal? 

2. If not, then how could I use Hugin to create a panorama image similar to "expected_panorama.png"? 

Many thanks for your help,

Alex

[wirz]

Hello Alex,

It looks to me like your final pano is simply frame-1, which suggests
some basic user error.

As for what is possible, do I understand correctly that we are looking
down on a non-flat roof and the left four and the right five rows of
panels are not coplanar? In that case, a distortion free panorama isn't
possible. For purely geometric reasons, hugin can only deal with two
situations: 1) all photos are taken from the same spot and the scene is
arbitrary, 2) the object is 2-dimensional and the photos are taken from
arbitrary positions. It can't deal with a moving drone and a non-flat
roof. However, you should be able to create two separate panos of the
two sides of the roof that are 2-dimensional as isolated objects.

cheers, lukas wirz

[Alex]

Hi Lukas,

Could you please clarify why Hugin is unable to handle a moving drone and a non-flat roof?

When using other panorama stitching tools, such as the Python Stitcher class, I am able to create a panorama image, as shown in the attached "panorama-Python-Stitcher.jpg." Please note that the result is not just the "frame-1."

Therefore, I assume that it is still possible to stitch all my images into a single panorama, covering the entire scene (in my images). Please correct me if I am mistaken.

I am reaching out for guidance on the settings or directions in Hugin that could help me to achieve my goal of stitching all my images into a single panorama, ensuring all areas of the scene are covered and overlapping sections are merged.

Thank you in advance for your kind assistance.

Best Regards,
Alex

lukas wirz 在 2025年2月11日 星期二晚上10:09:06 [UTC+8] 的信中寫道：

[wirz]

Hello Alex,

Starting with the general question: For a general pano, the condition
is that all images are takes from exactly the same spot. If the
position changed between two images, you might see an object from one
side in one, and from a different side in another image. And objects
may swap their relative position (think about trees in a forest). There
is no reasonable/consistent way to merge such images. Then there is the
corner case of a pure 2D scene where such perspective issues don't occur
-- then the camera may move.

Then there are cases like your's, where the scene is not too far from
2D, and therefore a not too far from acceptable result can be obtained
with a moving camera. But a strictly distortion free pano is impossible.

I gave this a try, see the attached pto. The result may be acceptable
depending on what you want to do with it. It's certainly much better
than your reference.

The steps were to manually place all control points, because the control
point finder fails on grid like structures (I think that went wrong in
your attempt: There were many wrong control points and all images were
optimised to the same position.). There are a few verticals and
horizontals. I placed all points on the left panel array and optimised
translation. One could improve it further with more control points and
a few more horizontal lines. One could also repeat the same for the
right half to get a good version of that.

best regards, lukas wirz

[Alex]

Hi Lukas, 

I sincerely appreciate your great assistance.

I have reviewed the Hugin project file you shared, "frame_0001-frame_0010.pto," and I must say it’s quite impressive. It has almost solved the problem I’ve been working on.

After thoroughly going through it, I have a couple of questions that I hope you could help clarify:
 1. Why do we need to specify a control point between frame_0001.jpg and itself (frame_0001.jpg)?
 2. Why is it necessary to specify 3 control points between frame_0010.jpg and frame_0001.jpg that form a circuit (from end to beginning), especially when, according to my drone’s flight data, these two images have no overlapping sections?

Thank you very much in advance for your kind assistance.

Best Regards,
Alex

lukas wirz 在 2025年2月16日 星期日凌晨3:24:03 [UTC+8] 的信中寫道：

[wirz]

Hello Alex,

> After thoroughly going through it, I have a couple of questions that I hope
> you could help clarify:

> 1. Why do we need to specify a control point between *frame_0001.jpg *and
> itself (*frame_0001.jpg*)?
> 2. Why is it necessary to specify 3 control points between *frame_0010.jpg*
> and *frame_0001.jpg* that form a circuit (from end to beginning),

> especially when, according to my drone’s flight data, these two images have
> no overlapping sections?

In both cases those are special control points that signify either
vertical or horizontal lines (with respect to the image, not with
respect to the 3D object). In frame01, I've marked one horizontal line,
and between the first and last frame there are three vertical lines.
Having one of them helps align the output image, while having several of
them helps warping it correctly.

> Thank you very much in advance for your kind assistance.

You're welcome!

cheers, lukas wirz