1

I have been tasked with simulating high speed thermal imaging, the background of this is that FLIR thermal camera operates at 9fps. The idea here is that the color camera which is already operating at a high speed (30 or 60 fps) can be overlaid with the thermal image with weighting given to each image (example: x for thermal pixel and (1-x) for color image pixel and then summing the two) - in other words, the color images would use the thermal image as an intensity map, if that makes sense.

For the sake of this discussion, assume that

  • There is no perspective correction needed between the two cameras
  • The thermal image and the color camera image are of the same size
  • We are only dealing with grayscale images from the two cameras, so the calculations are all in grayscale. The resulting grayscale image would be colormapped for final display, just like any typical thermal image.

However, I am unclear about the whole thing, and several questions pop as I think about it. AFAIK, what I am really looking for is a frame rate booster, but for thermal imaging. Like a predictor/interpolator. Agree? Overlaying or Intensity mapping is not applicable for video content with varying frame rates.

Does this idea make sense to anyone? Are there any known solutions/papers/implementations?

Thanks a lot!

chimp45
  • 67
  • 7
  • how is 30 or 60fps high speed for colour cameras? a few lines later you are talking about 2 grayscale cameras so no colour as mentioned befor? why not use a high speed thermal camera? they go up to several houndred FPS. – Piglet Oct 10 '16 at 21:54
  • You are right. But high-speed in my question applies to thermal camera (30 or 60 fps is high speed for a camera marked to function at 9fps). I had included a link explaining why FLIR limits to 9fps. Here it is, again... http://www.flir.com/cvs/cores/knowledgebase/index.cfm?CFTREEITEMKEY=353&view=70711 – chimp45 Oct 10 '16 at 22:31
  • What are you viewing with the cameras (Fire, plasma arc, just some solid)? what temperature interval ? add samples. Hi speed fps is something different you want just boost your low fps not use hi-fps (which for cameras are above 100 fps like 120,240 ... does not matter if RGB or thermal the hi-fps therm means special image recovery from chip usually by parallel interfaces in its architecture). – Spektre Oct 11 '16 at 08:04
  • Kind of like the inverse of chroma-downsampling maybe? – Mark Setchell Oct 11 '16 at 08:24
  • @Mark Setchell, I don't see the connection with chroma down sampling (where you are just weighting the luminance and chroma, here I am weighting two grayscale images). Please elaborate. – chimp45 Oct 11 '16 at 23:25
  • In chroma downsampling, you omit the chroma information in alternate (or some other percentage) of the pixels to save bandwidth in transmission and then regenerate the chroma by interpolation (or other method) to recreate the missing data at the other end. Here you have one slow source and one fast one and you want to in-fill frames in the slower stream to give the two the same framerate. – Mark Setchell Oct 12 '16 at 10:16
  • Yeah - but this is not interpolation from samples within the same image/signal. This is filling in based on content found in another image. Agreed - both are about filling in the blanks, but this is somewhat different. – chimp45 Oct 12 '16 at 15:55

1 Answers1

1

You can use visible range camera to interpolate/predict between/from thermal frames only for known physical process for example like:

  1. black body thermal light emission

    see B-V index and its correspondence to temperature. in the link there. But you need color feed for this. And also should compensate known surface color and Light scattering effects.

  2. fluid viscosity

    You can estimate fluid viscosity (which is strongly dependent on temperature) from gray-scale feed simply by measuring flow speed in known or computed volume against gravity or known actuator force.

  3. bi-metal deformation

    bi-metal deformation is also dependent on temperature which should be easily detectable on gray scale feed.

Community
  • 1
  • 1
Spektre
  • 49,595
  • 11
  • 110
  • 380
  • Spektre, is this interpolation/prediction more applicable for slow moving processes (based on the three you have mentioned)? The BV index seems a useful resource. – chimp45 Oct 11 '16 at 18:56
  • Sorry - Pressed a little too soon. Now completing - Spektre, is this interpolation/prediction more applicable for slow moving processes (based on the three you have mentioned)? Are there any useful resources/code discussing such mechanisms? The BV index seems a useful resource. I will look into it. – chimp45 Oct 11 '16 at 18:57
  • @sdla-45 fps does not matter really (as long as you got fast enough CPU) you can not do this for arbitrary image. You need to detect usable feature in both RGB and Temp feeds that are affected by temperature so it can be used as map for the interpolated images. After identifying feature you measure the properties you need to estimate temperature. It is highly unlikely you will find any relevant research for this as each implementation is specific for each case/task. – Spektre Oct 11 '16 at 20:31
  • @sdla-45 The approach you are mentioning is just morphing the images (by cross-referencing points of interest from both feeds) but that is not physically correct and resulting temp feed will have no valid information added (just visually smoother at cost of accuracy). – Spektre Oct 11 '16 at 20:33
  • Makes sense, I will put together an implementation and see how it goes. Picking targets would be a challenge. – chimp45 Oct 11 '16 at 23:13
  • Spektre - I am seeing the wisdom here... might not be what you are professing - but some wisdom. Basically, using the method I have described, I have extracted a lot of image detail from the luminance channel, which the low resolution thermal image lacks. Makes for a much crisper image. This applies well for all the use cases you have specified - for video content without a lot of motion activity. – chimp45 Oct 19 '16 at 17:49