Public Lab Research note

A DIY gas-finding camera?

by mathew | November 21, 2014 04:26 21 Nov 04:26 | #11377 | #11377

What I want to do

Explore whether its possible to make a camera that can see VOCs, like the very expensive FLIR Gasfinder 320 camera can. Tons of people at the Barnraising and at Science for Action conference wanted one. Me too.

cover image above from wikimedia commons

Here is a Gasfinder 320 filming a 240x320 pixel image of an oil train taken by the Vancouver Action Network. Note the VOCs pluming off the car.

Real cool stuff. Too bad rental was several thousand dollars for a weekend, and the cameras cost tens of thousands of dollars, up to $84,000 new (Excel file from Texas state aquisitions dept).

My hypothesis is that a camera of roughly half the resolution could be constructed using the techniques of early single-sensor mechanical television systems from the 1920's and 1930's and single photodiode sensitive in the 3200-3400nm range of the FLIR Gasfinder 320.

My attempt and results

How did mechanical televisions work? by scanning a picture one line at a time, so that the signal of a single sensor only represents a single point in the image at any given time. A series ofscanning systems, from belts with holes in them to mechanical disks and spinning mirror screws were tried, mechanical disks were the simplest system. Most had 30-60 scan lines, but lots of places on the internet list 120-200 lines as the upper limit of a mechanical scanning (Nipkow) disk. The Gasfinder 320 has only 240 scan lines. A mechanical TV camera works exactly the same way as the TV, except a lightbulb is replaced witha light sensor. Watch this great documentary on the Baird system starting at :57:

Questions and next steps

DIY projects abound for mechanical televisions that can be made from simple materials, but to use this scanning disk system with a camera, some sort of lens must be added, and a device for focusing captured light the right type of sensor.

We would need to:

  • identify a photodiode that works in the right infrared range (3200-3400nm);
  • identify lenses that also work;
  • filter light to the specific range
  • build a scanning disk system;
  • find or create software and hardware for converting the signal to a digital video source

the scanning disk system looks like it will be easy to build and can be bought off the shelf. The software doesn't seem too hard either. The hard part seems to be lenses and sensors. I think that some digital processing will reduce the flicker and produce much clearer images than the early mechanical TV sets.

Sensors & Optics

According to the FLIR website, the Gasfinder 320 operates in the 3200-3400nm range using an Indium Antimonide (InSb) sensor. According to Wikipedia this sensor is fairly high speed. More on the science of why it sees hydrocarbon gasses is here. To quote: "A wave number of 3019/cm (3.3-µm wavelength) relates to the absorption band of the antisymmetrical valence oscillation (a), whereas the deformation oscillation (d), whose movement is similar to the unfolding of an umbrella, causes a strong attenuation around a wavelength of 7.6 µm (at a wave number of 1306/cm). Both areas are suited for remote methane gas sniffing, with the stronger attenuation value favoring the use of the MWIR band around 3.3 µm. "

I can't find a InSb photodiode for sale, but they are out there.

I found some lenses for sale that also filter to Midwave IR, but they are crazy expensive, $300-$800. There seem to be issues with focal length and "chromatic" aberation (in the IR). these lenses seem to have the tightest filtration.

Television Kits & Resources

a couple DIY resources that are awesome:

a 30-line Baird Televisor kit for £29.95. It also comes with software for sending "audio pictures" to the Baird Televisor.

A printable PDF 30-line of a Nipkow disk.

OpenSCAD code for generating a disk of any size or density

''' // Parametric Nipkow Disk Generator // Andrew Davie

$fn = 32;                     // smooth curves (bigger=smoother)

DISC_DIAMETER = 150;         // in mm

SCANLINES = 32;               // number of holes!

HOLE_EDGE_OFFSET = 20;         // mm in from outer edge for first hole
HOLE_DIAMETER = 0.5;         // size of scanline pinholes
HOLE_SPACING = 0.6;            // spacing between scanline holes
DISC_THICKNESS = 1;            // mm

MOUNT_HOLES = 5;               // number of holes for mounting screws
MOUNT_HOLE_DIAMETER = 3;      // mm screws
MOUNT_CENTER_HOLE = 10;         // radius of center hole
MOUNT_HOLE_POSITION = 15;      // radius of mounting hole position


   // Start with the basic disc

   // ... and remove the following items...

   // Internal shaft/mount hole

   // Screw mount holes
   for (i=[0:MOUNT_HOLES-1]) {

   // Scanline holes around edge in spiral
   for (i=[0:SCANLINES-1]) {



Hackaday,hackaday project link

An alternative: Mirror arrays:

Why I'm interested

Even at low resolution and a $1000 of lenses and sensors, a mechanical TV camera would still be amazingly accessible compared to existing commercial systems, and immediately valuable in monitoring oil and gas sites.

As a mechanical device, most of the complicated parts can be laser cut or 3D printed and assembled by users. The design is not black boxed at all and is easy to modify and understand.


The mirror arrays are looking more and more attractive because of their light-efficiency and compactness. Also, since I can now order cast and high-gloss polished 3D printed sterling silver and gold-plated brass from Shapeways. it seems like a small, high-efficiency, optical grade mirror screw could cost a little over $100 to manufacture.

Here's the original mirror screw patent from 1928.. Check out these pretty good images from a contemporary re-built mirror screw set that has a ~7" picture. Here's some more info.

The original 1928 patent it is referenced by a 1955 patent using a mirror screw in a single-sensor electronic camera. This is almost exactly what we want to do, although its a slit-scan system for aerial photography and we want to take full frame images. It makes me hopeful that mirror screws were more than just a 1930's curiosity, and that the light capture of a mirror screw is enough to resolve an image.

Reply to this comment...

Can we digitize a 240x320 analog video signal using an standard audio jack?

A mac laptop captures a maximum of two cannels (stereo) at 96khz, with a 24-bit samplng rate on its internal soundcard. is that good enough?

240x320= 76800 pixels, fewer than the 96khz samples. I doubt that 24 bits of depth are required-- 8 or 16 bits per pixel would probably do fine. So a one frame per second image is possible.

a stereo signal could carry a synch signal (indicating where each line ends) in one channel and a video signal in the other. It seems a laptop should be capable of transmitting and digitizing a low-fi, low-speed video signal! that is great news, as very little electronics, beyond amplification, are therefore required to generate a video signal out of a mechanical TV, and all the processing can be performed in software.

Is this a question? Click here to post it to the Questions page.

Reply to this comment...

Finding a sensor that can do high-speed sampling in the range required is looking pretty expensive. Thor Labs has a series of amplified, cooled sensors in the right range-- for almost $4,000.

This chart from Hamamatsu is helpful. Indium Galium Arsenide (InGaAs), Lead Selenide (PbS) and Indium Arsenide (InAs) seem to be the high-sensitivity options. All of those sensors seem to require cooling.


Is this a question? Click here to post it to the Questions page.

Reply to this comment...

InAs (Indium Arsenide) looks almost perfect, with a high sensitivity and a peak sensitivity at 3400nm (3.4micrometers) when cooled to 196K. 196K is reachable with dry ice, which sublimates at -78.5 C, which is 194K. Dry ice cooling has three major advantages over other strategies-- it's cheap, it doesn't draw electricity, and dry ice clouds look really cool.

Here are a list of suppliers I found:

Reply to this comment...

While methane has a main absorption band between 3200-3400nm, it also has weaker bands around 2300nm and 1650nm. For the 3200-3400nm band EOC sells the Lms36PD series photodiodes. In the 2300um range they sell the Lms24PD.

Electro Optical Components also sells an evaluation kit for experimenting:

"for first-time users we announce sample systems and kits that enable fast preliminary experiments with mid-infrared LED-PD optopairs for different detection purposes."

It's described 'low-cost', but compared to what?

Is this a question? Click here to post it to the Questions page.

Reply to this comment...

Those lower wavelength bands may be cheaper or possibly not require as much cooling to get a fast sensor response. Good thinking.

We need a ~100khz response rate to photograph one 240x320 frame every second. That's the biggest sensor constraint, I think.

Reply to this comment...

You might appreciate what these guys did.

1:45 - they built a portable lab to monitor gasses around fracking sites.

Reply to this comment...

Here is the link to the full talk:

Is this a question? Click here to post it to the Questions page.

Reply to this comment...

Here is some alternative tech to the FLIR-- hyperspectral camera for methane visualization (which is one of the most expensive gases to monitor). May be of some interest?

Is this a question? Click here to post it to the Questions page.

Reply to this comment...

Login to comment.