I have a thermal camera that I would like to overlay automatically to the output of a RGBD camera (e.g., Kinect).

I can retrieve the intrinsic parameters of the thermal camera by using a special board I built, that allows the thermal camera to give as output a series of points similar to the ones obtained with a visible sensor and a typical calibration chessboard.

Once the intrinsic parameters of the thermal camera are obtained, what would the next steps be to align all the cameras? It's important to take into account that the two cameras deeply differ from each other, because they operate in different wavelengths.

  • $\begingroup$ @Raphael, with about 60 questions tagged computer-vision, I don't know if we need a more specific tag -- maybe others will have an opinion or suggestion. $\endgroup$ – D.W. Aug 20 '15 at 21:00

I suggest that you obtain both the intrinsic and extrinsic parameters for both cameras. The extrinsic parameters will include information like the offset/distance between the two cameras and the relationship between their two headings, which is needed to relate their images. (If only the relative position and orientation/heading of the two cameras is known, that will be enough.)

Once you have these parameters, you'll be able to map each pixel in the RGBD image to its corresponding pixel in the thermal image. This will let you align those two images. Then, depending upon the specific setup, you might need to do fine-grained image stabilization (e.g., if they are not on tripods).

It's not clear why the different wavelengths would have any impact on alignment of the two images.

See also https://en.wikipedia.org/wiki/Camera_resectioning.

| cite | improve this answer | |
  • $\begingroup$ Thank you for your answer D.W. and sorry for my late answer - The project was on hold for a while. After I finally retrieve intrinsic and extrinsic parameters of both cameras, as you suggest, I can map from the depth image to the thermal image, by using the rotation matrix and translation matrix between the two cameras. But, as of now, I don't have such matrices. I understand that for the alignment I can use the depth information, but how can I find the rotation matrix and translation matrix between the two cameras? $\endgroup$ – nibbana Oct 9 '15 at 13:55
  • $\begingroup$ @user1823280, please don't use comments to ask a new question. You should post a new question, if you have a new question. Thank you! $\endgroup$ – D.W. Oct 9 '15 at 15:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.