Building a thermal security camera

[lordratner]

Hello everyone! I tried searching first, but not a whole lot came up.

It's been a while since I used Bud's firmware for unlocking my E4, and it's still going strong. But it's also got me thinking about a new project I'd like to tackle, building a thermal IP camera.

I'd like a resolution of at least 320x240, matching the E4. I certainly wouldn't mind an even better resolution, but that would be cost dependent. The problem with the E4 is the Field of View, which is not great for a security camera.

Most of the DIY projects use 32x32 sensors or smaller. There are all sorts of ~320x240 devices available on eBay, but not a lot that I see with the appropriate lens. The second consideration is the video output. I need something that will play nice with a Raspberry Pi or ESP32.

Am I stuck looking for something functional with the correct lens, or can I look for two units, one with a broken sensor and the correct lens, then scavenge it to replace the lends on something more cheaply available like the sensors in the consumer devices (like the E4)

I don't know much about IR optics, so I don't know how realistic it is to replace the lens. But I suspect it will be much cheaper if possible, using multple damaged units to create a working one.

Anyone have experience with this type of project? Am I off my rocker?

Thanks,
Seth

[Bill W]

You could try scavenging old fire service cameras.

Around USD 200 should get you a 320x240 analogue video out 50° FoV, and the right models have removable stand alone cores. 
Not sure how that would play with a R-Pi, I suppose worst case an analogue to network converter.

Either keep the fire sealed box or rebox.  You'll either have a sealed lens or a sealable window.

Unless you knew what you were doing I would avoid swapping a lens for something much different.  You are likely to get miscalibration issues, even if it is just fall off to the edges.

Some scavenging ideas on my website, sorry no sales to USA.

[hap2001]

OP, I understand your suffering because I have the same problem looking for wide lens for thermal cameras: https://www.eevblog.com/forum/thermal-imaging/very-wide-angle-lens-for-tau/
The fact is most people want narrow lens to have higher spatial resolution, and most thermal cameras do not have wide FOV.
If you are lucky enough you can find an off-the-shelf IP thermal camera with ~90 HFOV on ebay.
Firefighting cameras are good choices if 50-60 HFOV is enough for you. Otherwise look at Flir Boson or Lepton that offer 95 HFOV, or changing lens yourself.
I have many spare lens from broken cores, but finding a suitable one that offering both wide FOV and enough image circle is not easy.

[DaJMasta]

I know it's for 'security', but what's your actual application?

Thermal security cameras exist as commercial products, usually in pan and tilt turrets, and you can get them on the used market - probably the simplest option.  The moving turret helps make up for the lack of spatial resolution by letting you choose where you point.  Making your own is a possibility, and while changing lenses really isn't a huge issue so long as you have the same mounting threads, it's difficult enough/expensive enough to find individual lenses that it's probably going to be cheapest just to find the lens you want on the core you want.

But again, the application is important.  You don't get much resolution even with a high resolution core and a narrow field of view lens, and even if you did, a recording of someone's face in thermal wouldn't go far for recognizing them or reporting it to the police.  A thermal camera will see heat in pitch black no problem, so a wide angle lens can be great for object detection over a wide area, but even with a 1024x768 core (some of the highest resolution available in the five digit priceranges), a wide angle lens is going to leave your complete target spotted at just a few pixels wide and nearly impossible to recognize.  If you actually want detail on your subject, you need a higher focal length lens and a smaller field of view, which would mean a best case for a thermal camera used for recognition of a subject would be looking down a long hallway or similar, but again, actually getting identifying information from your image when no one has thermal vision is tricky.

I think the case where thermal cameras get used for security is for dark areas with staff on site.  It's not that valuable to have a recording of a thermal image after the fact, but if you have a security team that can spot a subject on a thermal camera and then go investigate, it could actually be a useful tool, and you see thermal cameras used for search operations (land, air, and sea) for this very reason.  My guess is such an operation would have a fair bit of budget to spare, though.

[lordratner]

Thanks for the thoughtful response.

I'm not looking for identification, and most of what you said is inline with what I'm after. I'd like to use a thermal camera as a sort of screening mechanism. A lot of the newer AI stuff, or ML, or whatever abbreviation you want to use, has the problem of false positives. I'd like to have thermal for a sort of screening function. It would be great if I could have an image good enough to identify if the target is human, animal, or vehicle. My range is probably out to about 300 feet.

I haven't seen many used PTZ cameras, so I don't know what they go for on the used market. I wouldn't mind that as a solution, I would just have it set to sweep the area. If the cost can be driven low enough, I would also just consider making enough cameras to cover the area. A stationary camera has the advantage of being able to set up detection zones, which will reduce the false positives greatly.

Ultimately if I can either get something like a core that I integrate with my own controller, or an already-built solution that will output the feed over any common IP protocol (http, rtsp, etc), I can make it work. This project is still very much in the conceptual stage, but I've been pondering it for a while. I figure $1,000 is about as much as I want to spend on it. My E4 was $800 new back in 2019.

The easiest answer would be one of the FLIR Hadron cores, but those are very pricey.

Just idea shoping, and this is the best place for thermal advice. Thanks!

[DaJMasta]

The PTZ cameras I've seen are sometimes security and sometimes maritime variants, and generally have a tau/tau 2 core in them.  That will give you analog video out, which isn't great, but which can be digitized pretty easily on a USB device (something that supports UVD would probably be the easiest to interface with externally).  Digital output on a thermal camera outside of still frames still comes at quite a price premium.  IP would be the cheapest option for digital, but they tend towards their scientific line - I think the cheapest FLIR IP cameras I've seen on the used market have actually been for mammography screening - A20 variants if I remember right.

I think the dedicated security camera style will tend towards longer focal length lenses, which if you're using it as a screen for a higher resolution camera, could be useful if they have similar zooms, but if you want it to try and preselect areas of interest for a higher resolution/higher zoom camera, the wider angle lens and just going for the hot pixels could be a better approach.

There are also the newer cores mostly from Chinese manufacturers, and if you're able to get your hands on the digital data stream, that could be a very cost effective option with a variety of lenses to choose from, since they could probably be bought new.  I know some of those SDKs are locked down like the rest, but I've seen some threads here with people figuring out at least some of it, so it may be an option worth exploring.

[Bill W]

A shame the eBay Argus4 are so overpriced.  You'd get one new for some of those used prices if you asked the right people !

Sealed IP67 box
50° HFoV
320 x 240 (check model, some are 160x120)
RTSP (well maybe.... and only at 10Hz)

However the RTSP function was dropped several years back as the Ethernet bridge chip went obsolete and hardly anyone used the function so we left the circuit unpopulated.  That means the more recent ones won't stream and there is no easy way to tell from the outside for sure, there was not even a clean serial number step change.

I can't help unless a 9Hz camera would do for you, to be export control exempted as I don't have a license UK to US
Potentially the stream would be a bit clunky as you have a 10Hz send of an image that is 9 different 'pictures' per second.