To obtain the 3D information from a single image is a hard task. There are many algorithms for 2D to 3D image conversion, but good results are subjects to restrictive condtiions and for therman imaging, it would be probably harder.. But if you have serveral thermal images, you can infer 3D information through disparity estimation or Structure from Motion.

Hello José Luis Herrera Conejero,

thanks for answering me. i want to know that if we have several 2D thermal images how to transform through disparity estimation or Structure from Motion.

If you are looking for 3D image presentation like to express different intensity by its level, you can use "surf" representation.

hello Pavel Chmelar,

thanks for replying, please send me some reference on surf representation.

The first thing which you require would be to buy a 3D thermal camera....i dont know whether it is available in the market or not........

But from 2D thermal images to generate 3D images is a tough task.....it would be possible if you can get exact depth values of each (x,y) point corresponding to the temperatures captured during 2D thermal imaging.

For example here: http://stackoverflow.com/questions/16342164/how-to-plot-a-surface-with-a-texture-map

You'll use your image intensity as Z coordination. 

As far as I know there is no such thing as a 3D thermal camera. It would be quite expensive anyway. However, you could use a conventional 3D sensor (e.g. a Kinect or a stereo camera) together with the 2D thermal camera in a fixed arrangement with the cameras close to each other, looking parallel into the same direction. To get the mapping from pixels in the thermal image to the depth image you have to do a so-called registration of the two systems (see e.g. http://www.cse.oulu.fi/~dherrera/papers/2012-KinectJournal.pdf). This has to be done only once. After that you should have a fast 3D thermal camera system.

Good luck and best regards, Ralf

hello Ralf Vandenhouten

thank you much. If you have more reference please send a attachment.

good day

regard

Hi Rajesh,

the basic idea of registration is quite simple. From the depth image you can reconstruct the 3D object point cloud by reprojection. This is described in detail e.g. in the O'Reilly book "Learning OpenCV" (http://shop.oreilly.com/product/9780596516130.do). Once you've got the 3D point cloud you can project it to the direction of the thermal camera, i.e. you can calculate how the point cloud looks like from the point of view of the thermal camera. For this you need the exact position, orientation, and focal length of the thermal camera (and lens distortions, if any). The result is the "registered depth image" where each pixel is at the same position as the corresponding pixel in the thermal image, i.e. combining the two images yields a thermal-depth image. 

A little programming effort from your side will be necessary, but some existing open source frameworks may be helpful, particularly OpenCV and the Point Cloud Library (http://pointclouds.org).

Best regards, Ralf

Shareware has a list

http://thermal-imager-software.sharewarecentral.com/

All details per the above are accurate. Unless you need high registration of 2D thermal images to 3D structure obtained from active depth sensors or passive methods (e.g. structure from motion or stereo matching), you may adjust thermal images as texture to the 3D mesh of your structure. Accurate registration of your depth sensor to your thermal camera may require a specific calibration object to computer the respective pose of each cameras.