Is thermal right for me?

[wittaj]

Thanks for sharing your experience with thermal. Your comment led me down a rabbit hole on thermal sensors, and one thing I noticed was that the TPC124X-AI-S2 (maybe the same as TPC-DF1241-S2?), seems to have different advertised sensitivity vs the DH-TPC-DF1241-D3F4. Specifically the former is rated with a Noise Equivalent Temperature DifferenceI of 40mK while the latter is rated at 50mK. I wonder if that's meaningful, or reflective of a more broad difference. Coming back to your recommendation, I'm not sure if PTZ would be good in my specific case. I'm thinking the alternative would be adding IR emitters?



Thank you for the guidance! Because there's a porch my plan was actually to use the variofocals near the rear entries and windows for identify. Then, use additional cameras for overview to see further out. That's where I thought that maybe the Color4Ks could fit that overview role. But once I saw the image at 16ms, I knew it wasn't going to work out.

I've read (and reread!) your prior review of the TPC124X-AI-S2 and had a couple questions. Wisdom here seems to be that you can't really trust DORI distances on spec sheets. That makes sense. But you've referenced being able to trigger on a person at over 400 feet with the 3.5mm version while the spec sheet says it can do something like 479 feet. Does that mean that the thermal performance is close to what's on paper? Related- the thermal perimeter distance for a human is specified as a completely different number - 80 feet. That appears to be very different vs all of your testing, but I'm not sure what the difference is. What am I missing?

Yeah, in my testing, the Detection and Observe numbers provided for thermal cams seems to be pretty accurate. I think because the image looks the same regardless of the light conditions.

Now the Recognize and Identify for thermal, no way LOL. Here is someone less than 10 feet from the camera or well within the IDENTIFY distance for the thermal LOL:



The perimeter distance number you reference of 80 feet is under the "General" number, so I am assuming that means they distance they have high confidence in that it would trigger for anyone within that distance.

But you do have to manage your expectations - when I referenced the 400 feet - I could tell that was a person and not a deer as an example, but would never be able to provide any additional details like male or female, etc. but I was able to see where the person was going and hiding behind bushes that I simply could not see at that distance with a visible camera with a shutter speed set to minimize blur.

 

[Dis7]

Gotcha, I did laugh when I read the Identify stuff on the product sheets LOL

 

[luk8899]

At night do the following:

• Make it 1/60 (16.67ms) and post the same field of view - I bet it will be a lot darker.
• Then make it 1/120 (8.33ms) and post the same field of view. It should be even darker
• Then make it 1/10,000 and that image should be completely black - if it isn't then you know the firmware is playing with the image and at that point it doesn't matter how good the sensor is in the camera.

Sadly, the slider in the camera web intetrface goes from 1/1000 to 1/100 and then immediately to 1/10 (but has hundreds of settings between 1/1000 and 1/10000) So I was able to test only at 1/100 and 1/10. Based on my testing I believe the Auto setting is somewhere around 1/15~1/30. The image at 1/100 was too dark, a human could be spotted, but barely so. However, today is almost a half moon. So not an ideal night for such testing. Still dark enough for me not to see my cat sitting in the road and I had to stop running or I could've tripped.

When I have more time I'll play with the cgi interface. This will let me set the shutter to whatever number I want I think. The image is indeed completely black on the fastest shutter setting.

On Auto setting, when I run in front of the camera and even walk briskly at near range the face is far too blurry for night identification. (Good I haven't bought the camera for this - I would be pretty dissapointed). At higher distance and when movement is slower it is passable. For my purposes this is perfectly fine (as a part of a larger setup with PIR sensors, additional cameras etc).

However, I discovered something else. There is something very interesting going on with the IR illuminator when I'm zooming. It seems it zooms the IR too. See the video below. First is the "running in front of the camera" test. Then I'm chasing my cat with the zoom. Then, the same scenes are shown as seen by very wide angled 2.7mm lens imx415 camera. You can clearly see the illumination beam changes shape when I'm zooming on to the cat. On the video it looks as if he can see the spotlight, but I assure you all you can see is a very faint red glow in the distance.
Edit, this is 4K, I recommend to watch in full screen in 4K.


Also, regarding the thermal imagers. Especially looking at stills it seems 256x192 is not a lot of resolution, but at a nice FPS one can really make out a lot more detail.

 

[wittaj]

OK that is a good test and appears it has some decent firmware.

At a 1/100 shutter, there would be lots of dark places that a person could be and not seen whereas the thermal would pick them up.

You will find at night that EVERY camera on auto setting will result in blurry motion. They just slow the shutters down too much. That is why we set them to manual and no slower than 1/60 but go as fast as we can given the light we have.

Some varifocal cameras will have a zoom priority on the infrared, so we are probably seeing that, but we are also seeing a little bit of the 2.8mm camera adjusting the gain based on the available light that can also give an appearance of the infrared changing shape.

 

[bigredfish]

The IR seems to be pretty good. Maybe a Zoom Priority type function .

Some if the big Dahuas like the one I showed have “laser IR”, a very clear circle based on how you set the degrees. Auto it’s wide open.

 

[luk8899]

For anyone interested as we're also talking about nigh performance of visual imaging sensors here. I checked the Dahua cam from 2019 with the 1/1.2 4MP sensor (IPC-Color4K-X, IPC-HFW5849T1-ASE-LE) and here in the EU it still costs $330. Ouch! This is pretty much what I paid for the thermal dahua. Having watched the "world's first" reviews, it is a very nice camera. But for $330 in 2024? This is about a cost of 5 cheap no-ptz imx415 cameras, or two 30x optical varifocal zoom ones. So, how about the 4MP version then? It is cheaper, but still about $250 here.