Jump to content
UltravioletPhotography

LWIR panoramas, tech notes and a photo


Recommended Posts

Stock E60, and I explained in the top post (first one on last page) how I made it from the RAW, and that’s why there are no logos. The 16 bit RAW is stored inside the jpg’s and exiftool can extract it.
Link to comment
Extract the RAW from each FLIR jpg with exiftool. Use PS to adjust the histogram and resave all images in a batch using a custom Action. Use PS to stitch.
Link to comment

I see a few FLIR E60 completed listings on eBay starting at even $650 and up... so that is even less expensive than the E4/8.

And some E30 to E60 mods, etc., which might appeal to me more than a stock E60, simply because I would not need to use exiftool to remove logos.

They also sell mods the E30/40/50 for $200, if you can find one of those that is cost effective.

https://www.ebay.com...LIR+E60&_sop=15

 

Yeah, so no need to explain the exiftool procedure.

 

I don't know if the E30/40/50 upgrade to E60 mremoves the FLIR logo as with the E4/5/6 to E8 upgrade.

I will try to find out.

Link to comment

I am sure governments spent untold amounts of money on developing radar.

Now we have one in every home, available for very little cost.

 

well.. maybe not the home but marine pleasure craft certainly benefits.... don't you have a radar on your boat??

Link to comment

This is what I was referring to:

"The development of the cavity magnetron in the UK made possible the production of electromagnetic waves of a small enough wavelength (microwaves). American engineer Percy Spencer is generally credited with inventing the modern microwave oven after World War II from radar technology developed during the war. Named the "Radarange","

 

https://en.wikipedia.org/wiki/Microwave_oven

Link to comment

Cool... but I am still not clear how the logo is removed. Can you explain this to someone who has never done it before?

 

exiftool -b -RawThermalImage IR_314159.jpg > IR_314159.tiff

 

Even if you get a hacked E60, you probably STILL ought to use exiftool to extract the RAW, because the quality will be higher than the JPEG.

Link to comment

Andy, please explain to me how you do the FLIR logo removal.

I don't think the mod available for E30/40/50 to E60 removes the logo like it does with E4/5/6 to E8.

I opened a few of my FLIR JPG files in exiftool and didn't see anything that looked like a logo switch.

Maybe you can explain the process. Thanks.

Link to comment
Cadmium, I just did! The logo is not part of the RAW! Just do what I did above (post 33) with the exiftool and you will see.
Link to comment
Oops, no, sorry, but I'm not that excited, I think my writing style tends toward "excitable-sounding," though. Just divide the apparent excitement level by 2 to get the calibrated excitement level.
Link to comment
I get that response from people sometimes also. I'm always HAPPY!! and exCITed about stuff!!!!
Link to comment

Oops, no, sorry, but I'm not that excited, I think my writing style tends toward "excitable-sounding," though. Just divide the apparent excitement level by 2 to get the calibrated excitement level.

 

Yes, party animals, one and all!!! :excl: ;)

Sorry if I misinterpreted your content and punctuation.

 

Andy, as I replied to your private message, your message looked exasperated with me for asking the question 'after' you posted the answer...

As I said, we wrote those messages at the same time, so I didn't read your 'answer' until after I asked my question.

Mind boggling, I know, but the exclamation mark(s) in that message seemed to imply that you were, shall we say, annoyed, astonished, or otherwise bewildered with me asking after your answer.

No big deal, messages cross paths like that sometimes, it happens, just pointing out the times they were written so you know why I asked after you answered.

 

Thanks for your info and instructions, I will try it out sometime soon.

Sorry again for thinking you were frustrated with my time delayed question.

Link to comment

Yeah, I abused the exclamation points a bit recklessly in this case. No matter, I think we're all straightened out now.

--

Have a picture of my local courthouse.

post-94-0-57748300-1516594182.jpg

 

ETA: the haunted mansion

post-94-0-05766800-1516596695.jpg

Link to comment
  • 1 year later...
Ultrapurplepix

I thought I'd add my couple of cents' worth on LWIR panoramas. I have been making these for a number of years in various locations around the world. Most of the time I have used Therm-App brand cameras: the early images were made with the 384x288 resolution original camera; later ones with the more-sensitive 640x480 pixel Therm-App Pro. Images are sometimes stitched manually but I mainly use Microsoft Image Composite Editor.

 

There are various types of thermal camera, not only in terms of spectral response range that was mentioned in an earlier post, but 'cooled' and 'uncooled' cameras. Briefly, cooled cameras use an expensive, mechanical (limited life) helium-based cryogenic cooling system for the sensor, but give fabulous, low-noise images. Uncooled cameras are at least an order of magnitude cheaper, use a different type of sensor and are less sensitive (noisier). The optics for a cooled LWIR camera are typically f/4.5; for an uncooled camera, f/0.7 to f/1.4 are about the practical limits, with f/1 or f/1.2 being the most common. I have no experience of MWIR or SWIR cameras but I suspect they follow a comparable pattern.

 

Most thermal imaging camera firmware / driver software strives to give a good image irrespective of what you're looking at, so various kinds of automatic gain control, gamma manipulation and other enhancements are going on all the time. It can therefore be quite difficult to get the same exposure settings across a widely-varying scene. I'll come back to this topic. Many cameras overlay a logo that you can't remove (I'm looking at you, FLIR), which makes compositing that bit harder. Finally, uncooled thermal cameras are actually quite unstable, with per-pixel drifts. In fact, someone memorably once described a thermal camera as "an array of pixels with wildly different responses", which are 'tamed' by applying gain and offset calculations on a per-pixel basis (and there are further global corrections as well for things like ambient temperature). And then you have to do the processing to come up with a usable image...

 

Some cameras can achieve pseudo-stable operation by careful compensation and / or the use of a 'shutter' that periodically interrupts the view of the scene, close to the sensor and presents all pixels with a uniform, unfocused reference, from which minute-by-minute correction parameters can be finessed. Some uncooled cameras also operate the sensor at a little above room temperature (say 35°C) using a Peltier device, carefully controlled to keep the temperature as constant as possible and thus remove or reduce one set of variables. Even so, thermal imaging is akin to magic - you're trying to image something at (say) 20°C through a lens that's 'shining' brightly at the same temperature, onto a sensor that itself is at a similar temperature. Each pixel receives only a miniscule amount of energy from the scene, sitting on top of a huge background level. The following may be technically in error but try to think of detecting a signal of a few nanovolts sitting on top of a couple of volts - and as already established, adjacent pixels could easily have 'quiescent' voltage levels several orders of magnitude different from the signal level.

 

And yet, with all these difficulties, you can easily buy a consumer-grade thermal camera with VGA resolution that's easily capable of detecting scene differences of 0.05°C. That's more than good enough to see veins and arteries under the skin, or (thanks to the varying emissitivity of different materials) see perfectly well in complete darkness. You'll pay in the region of US$3000-4000 for something with this capability. Lower resolution thermal imagers start at a few hundred dollars and used (often ex fire service) cameras offer very good price/performance ratios, but usually need a frame grabber to capture their video output.

 

Some third-party driver software exists, particularly for the Android phone-based 'dongle'-style cameras. My present favourite, for XTherm and Therm-App cameras, is ThermViewer, although for the latter camera ThermApp Plus has some better features such as lockable exposure.

 

 

 

Here is a small selection of my thermal work over recent years. All are linked to much larger versions on Flickr, where you'll also find a write-up on each image.

 

Liberation Square, St Helier, Jersey. Original is 9176 x 4362 (~40Mpix), from 640x480 tiles.

 

post-27-0-03129000-1558699441.jpg

 

 

Blood vessels in human leg. Original is 1280x960 (superresolution upsampled from 640x480).

 

post-27-0-35449000-1558699587.jpg

 

 

Wall mounted space heater. A palette experiment. Original is 1280x960 (superresolution upsampled from 640x480).

 

post-27-0-34269400-1558699700.jpg

 

 

Gibraltar and harbour. Original is 6421x2886 (18Mpix), from 640x480 tiles.

 

post-27-0-74927800-1558699844.jpg

 

 

Wake in the Nile. Original is 384x288.

 

post-27-0-78134500-1558699973.jpg

 

 

Apologies to Warhol. Original images were 384x288. Shows Fuji IS Pro with Baader U filter. Can't remember the lens...

 

post-27-0-40697200-1558700106.jpg

 

 

Tower Bridge, London (unfinished). Manual tiling from 384x288 images. Main file is 5380x2179 (~14Mpix).

 

post-27-0-23961700-1558700305.jpg

 

 

Abu Simbel, Egypt. Original is 2176x1053, made from 384x288 tiles.

 

post-27-0-96298500-1558700421.jpg

 

 

Great Grey Owl. Original is cropped from 384x288 source. Visible image was probably by D700.

 

post-27-0-53302100-1558700537.jpg

 

 

Barbary macaque, Gibraltar. 640x480 original, superresolution upsized to 1280x960.

 

post-27-0-75542200-1558700662.jpg

 

 

Great Pyramid, Giza, Egypt. Original is 384x288, upsampled in Paint Shop Pro.

 

post-27-0-84668000-1558700859.jpg

 

 

Cathedral of Our Lady, Strasbourg, France.

 

post-27-0-92559400-1558701008.jpg

 

 

 

 

That's enough for now. I'm off for a coffee.

 

post-27-0-64064900-1558701125.jpg

Link to comment
Andy Perrin

Lovely pics, and this is confirming my theory that gothic architecture is particularly well-suited to thermal imaging. All the tiny nooks and crevices give rise to interesting thermal behavior. The cathedral is my favorite, followed by the macaque.

 

Some comments:

1) EXIFTOOL neatly removes the FLIR logos and allows access to the RAW temperature values as a TIFF.

2) Re lenses, SWIR lenses are much closer to ordinary lenses with different choice of glass and AR coating than they are to LWIR lenses. I don’t know what the situation is for MWIR.

 

Some questions:

1) What kind of frame grabber would you use? I have this problem for the SWIR camera and I can’t find anything over 10 bits that works with NTSC/RS-170 video.

 

2) How do you do the super resolution images?

Link to comment
Ultrapurplepix

1. I have tried a variety of frame grab methods ranging from the analogue inputs to a Mini-DV camcorder to standalone DVCRs that write to SD card (think dashcam with an analogue input) to the ubiquitous and cheap USB dongles that capture S-video if you ask them nicely. All basically worked as advertised, with the Mini-DV probably having the edge.

 

Whatever bit-depth of the sensor ADC (perhaps 12 or 14 bits), I suspect that most of the older generation cameras output only 8-bit video. I believe there's at least one ery old one that uses 6 bits. It's good enough for the originally intended use.

 

2. Superresolution is a technique that extracts better-than-sensor resolution by repeatedly fractionally offsetting the position of the sensor or lens or camera, detecting the difference between successive frames, and synthesising extra resolution data (pixels) from that. The offsetting can be as simple as normal camera shake, or as scientific as mounting the sensor on a moving platform that jiggles up and down and side to side under processor control.

 

The theory is that thermal pixels are not uniformly sensitive across their surface area, and the 'fill factor' (proportion of the sensor surface that contains usefully sensitive parts) is necessarily less than 1.

 

Imagine for a moment two adjacent pixels. Imagine further that only their central 50% is usefully sensitive. If you move the image half a pixel across, you can now detect energy (detail) that was previously lost to the gap between the sensitive areas.

 

In practice, fill factor is >66% but that still lends itself to a significant increase in effective resolution.

 

One can achieve a related effect using motion compensation calculations on moving images. There are several Windows packages such as Video Enhancer from incogniton.com and you'll also find a good description of their technique. http://www.infognition.com/articles/what_is_super_resolution.html

 

It is even possible to combine both techniques.

 

ThermViewer for the Therm-App and XTherm cameras includes (selectable) 2:1 wiggly-hand superresolution that works exceptionally well. 1280 x 960 thermal images at 25 frames per second requires a lot of processing power but even a slightly older phone (I use a Galaxy S7) is good enough for real-time rendering. The images look smashing and provide an excellent source for ultra high resolution panoramas.

 

I neglected to say, earlier, that good thermal cameras are expensive for good reason. They are fiendishly difficult to make, require complicated calibration and the sensors are remarkably large. Present state of the art is pixels on a 12um pitch (that's less than a wavelength of the light it's seeing). A Full HD sensor (about state of the art) is therefore a minimum of 12 x 1920 = 23mm by 12 x 1080 = 13mm, more or less. Packaging (which has to be hermetically sealed and high vacuum) doubles the size. Most sensors commonly available today have pixel spacing as large as 17um (the previous generation was 25um). A 17um Full HD sensor would be about 33mm x 18mm - roughly the size of a 35mm full sensor, but a fabrication nightmare. (When you read up on sensor design you learn that the actual thermally sensitive pixels are necessarily manufactured as a separate unit from the "read-out IC" (ROIC) and each pixel is individually soldered to corresponding contacts on the ROIC. That's two or four million solder joints per sensor; any failed joints give dead pixels. And then there's the lenses: glass is useless at long wavelengths because it's opaque. The most usual lens elements are made from exotic, high purity single crystal germanium, with each element individually turned on a diamond lathe.

 

It's a fascinating subject and I could go on for hours. And I'm not even an expert!

Link to comment
Andy Perrin
Whatever bit-depth of the sensor ADC (perhaps 12 or 14 bits), I suspect that most of the older generation cameras output only 8-bit video. I believe there's at least one very old one that uses 6 bits. It's good enough for the originally intended use.

Yeah, I'm looking at my SWIR camera, which is NTSC (so, bit depth not a concept for analog). The camera in question is non-standard in nearly every way, so who knows what it's doing internally. I just don't want to capture the results with an inadequate frame grabber. The thrust of the question was on frame grabbers for machine vision cameras, because that's what I'm dealing with.

 

There are several Windows packages such as Video Enhancer from incogniton.com and you'll also find a good description of their technique. http://www.infogniti...resolution.html

 

It is even possible to combine both techniques.

 

 

ThermViewer for the Therm-App and XTherm cameras includes (selectable) 2:1 wiggly-hand superresolution that works exceptionally well. 1280 x 960 thermal images at 25 frames per second requires a lot of processing power but even a slightly older phone (I use a Galaxy S7) is good enough for real-time rendering. The images look smashing.

Hm, maybe Video Enhancer would work for my SWIR stuff. I take it that ThermViewer only works on Android and only with XTherm cameras? Not just any video/images? I use MATLAB for most of my image processing, and I've considered writing my own superresolution code, or downloading one. That might be a good method in my case.

Link to comment
Ultrapurplepix

Yes, ThermViewer is an Android app that communicates directly with the camera via USB; in concept it's a bit like a 'Windows device driver with a user interface front end'. The camera interface is proprietary but there's an SDK, which is how come there are at least two third party driver/display apps.

 

I recommend you try the Infognition Video Enhancer 2. It has a generously long trial period, doesn't watermark, and basically 'just works'.

 

To see what it can do from a 640x480 thermal video, watch this video I shot at the Gibraltar/ Spain border.

 

 

(I recommend you download the video and use your favourite media player - look for the download symbol on the bottom right of the black image area - because Flickr tends to resize video to fit the window rather than letting you see it at 1:1).

 

At the time ThermViewer didn't yet support superresolution on the 640x480 camera (not surprising really, when you learn that the software author had never seen a Therm-App Pro and I had only just received mine - its serial number is 000001 so there weren't many around!)

Link to comment

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now
×
×
  • Create New...