Find the Focal Length

[ycuskun]

Hello,

I have reviewed many stitching works. But your work is very successful, I congratulate you for this.
I want to use it in my own work. But I don't know how to calculate the focal length value. Can you help with this?

[SSARCandy]

please refer to this
5096a28

[ycuskun]

How can I learn from the AutoStitch demo application?
Or how did you calculate it?

[SSARCandy]

you can try this logic:

[ycuskun]

I want to ask one last question;
Can you verify the variables in the image?
p:sensor width
w:object width
f: focal length (the value I need to find)
d: distance from object

Thank you.

[ghost]

Dear Colleagues,

First of all thank you @SSARCandy for your contribution... Your code is amazing: I ran the code for your sample datasets and it worked perfectly!

However, now I am trying to use it to generate an equirectangular panorama from my own data (Drone Imagery) and the focal length of my data in pixels seems to be too high… so the code is not responding.

I have tried several methods:
First, extracting the focal length from the image exif_metadata:

And second, by computing it with the following formula (knowing that my drone has a CMOS sensor of 1´y found the CCD width is 13.2 mm and the focal length in mm 28 (eq. 35mm):

In any case the resulting focal length values are greater than 10000 pixels and the code doesn’t work.
Could you pleas tell me what I am doing wrong?

Thank you very much for your help!

Greetings,
Sebastian

[SSARCandy]

hmm is it possible to point out where the code gets stuck?

[ycuskun]

I am also getting the same error as @Sebasmedd26. The code doesn't work when I try it with the pictures I took. The shift value is always zero. Sample output;

The part to the left of the stitched image is the first image I took. The part on the right is the fifth. Each new image is overwritten on the right.

[SSARCandy]

could you provide test data? I will try reproduce it.

[ycuskun]

panoramic_images.zip

I calculated the focal length as 1600. I calculated like @Sebasmedd26.

[ghost]

Hey, thank you for the fast reply. Can you give me please an E-Mail address to give you access to the dataset from my Google Drive? ycuskun ***@***.***> schrieb am Di. 19. Apr. 2022 um 13:10:

…

Uploading panoramic_images.zip… I calculated the focal length as 1600. I calculated like @Sebasmedd26 <https://github.com/Sebasmedd26>. — Reply to this email directly, view it on GitHub <#22 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ATQ3B3EKM343T6AYEL4NIJDVF2ICZANCNFSM5SYMO63A> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[ycuskun]

yasincuskun@gmail.com

[ghost]

@SSARCandy I cannot upload the dataset here. It is too heavy (251 mb in .zip).
@ycuskun I already share it with you on GoogleDrive.

Greetings!

[Ning-yuan118]

@SSARCandy I cannot upload the dataset here. It is too heavy (251 mb in .zip). @ycuskun I already share it with you on GoogleDrive.

Greetings!

I have the same problem as @ycuskun, please help me.

[Ning-yuan118]

I am also getting the same error as @Sebasmedd26. The code doesn't work when I try it with the pictures I took. The shift value is always zero. Sample output;

The part to the left of the stitched image is the first image I took. The part on the right is the fifth. Each new image is overwritten on the right.

Did you solve the problem, please help me!!!!

[SSARCandy]

could you check the environment? recommand

• python3.6
• opencv3

[Ning-yuan118]

could you check the environment? recommand

• python3.6
• opencv3

I checked the environment, but the shift was always zero. while i used autostitch, the perfomance looks well.

[Ning-yuan118]

could you check the environment? recommand

• python3.6
• opencv3

I don't know whether it's the problem with focal length. Is it a problem with my calculations?

[Ning-yuan118]

like this

[SSARCandy]

hmm I guess some bug in RANSAC algo

[Ning-yuan118]

thanks for your reply. when i used demo image, it works well. maybe my data is not suitable for the orignal parameters?

[Ning-yuan118]

my bug is like @ycuskun issue. did you give some solutions to him?

[SSARCandy]

RANSAC_THRES_DISTANCE = 3

try adjust this value

[Ning-yuan118]

I have changed the distance value to 500, but the result seems worse than the past. that's duing to the value focal_length.

[SSARCandy]

const.MATCHING_Y_RANGE can also be tune

[Ning-yuan118]

It still not work well, the image has a resolution of 2560 * 1440. Can it be a problem with image resolution?

[SSARCandy]

the matching algo works better if horizontal Y is not offset too much