Pulsar Accolade 2 LRF Pro <25mK 640x480???

[Ultrapurple]

Hi folks

I recently came across the not-cheap Pulsar Accolade 2 LRF Pro thermal monocular that claims a remarkable <25mK NETD from its 17µm 50Hz 640x480 uncooled LWIR sensor and 50mm f/1.2 germanium lens. Certainly, the few bits of video I've seen do suggest it's an excellent performer.

I can't afford the four to five thousand pounds it'd cost here but I wonder if anyone else has seen it, possibly even in the flesh? Any idea what sensor is in it?

And yes I do know that NETD figures can be massaged to the point of meaninglessness, even if there was a meaning there to begin with, but the pictures do look rather good, if heavily processed.

[mikeselectricstuff]

25mK, 50FPS, but not  at the same time...

[Ultrapurple]

I had an idle thought: I wonder if it's running the sensor at 100Hz and averaging frames to reduce the noise and up the effective sensitivity.

As they're apparently readily available in the UK and many other countries I guess they're not using American-made sensors. The Ulis /Lynred Pico640 (which, admittedly, has been around a while now) will run at up to 120Hz but the NETD is nowhere near 25mK - the primo-grade version of that sensor only guarantees <40mK at 30Hz with f/1 optics. But that's quite a long way from achieving 25mK with f/1.2 optics.

[Fraser]

The NETD of microbolometers is improving as the technology is refined…..


The Lynred PICO640 low noise “S” version with NETD <30mK (note the requirement on frame rate though)

https://www.lynred-usa.com/products/vga-resolution/pico640s-infrared-sensor.html


The standard version PIC640 Gen2 with NETD <40mk

https://www.lynred-usa.com/products/vga-resolution/pico640gen2-infrared-sensors.html


In addition to the improving noise levels of newer microbolometers, the OEM’s are using more effective noise processing in their software so the images presented to the user can be very clean and impressive. It is far easier to clean an image that is not going to be used for Radiometric thermography and rifle scopes etc are such an application where significant software image enhancement is often applied. This can result in an ‘overly processed’ appearance but, if done well, a pretty picture is produced. There is also the option to employ image stacking as the microbolometer is operated at a high frame rate. Such real time image stacking impacts displayed image update rates and can introduce motion blur but the noise content of the image can be significantly reduced.

You will see that there are two versions of the Pulsar Accolade 2 ?… a Pro and non-Pro version. I would suggest that the Pro version uses an enhanced ‘specially selected’ microbolometer such as the PICO640S and the non PRO version uses the standard, but still excellent PICO640 Gen 2. I have no confirmation of this however.

As a side note, I was very surprised at the low noise levels seen in images coming from the Dianyang Technology PCBA thermal analysis camera. That uses a Chinese made microbolometer and , though not perfect, that core produces impressive imagery from 12um pixels. I suspect a fair amount of image processing is employed but for many non Radiometric applications, the future is looking bright where image noise levels are concerned and the production of ‘pretty pictures’ is concerned  The Chinese microbolometer manufacturers  have also released microbolometers with tiny 10um pixels and 8um pixels are just starting to appear.

I suspect we will soon see a similar situation to mobile phone camera technology where the data captured by the lens and sensor array is far from optimal, yet looks fine to the user thanks to some very clever image processing. The image processing will obviously be different for a thermal sensor, but digital image enhancement gets better with every passing year.

Fraser

[Fraser]

Ultrapurple,

You mirrored my thoughts on the PICO640 but I think it will be the PICO640S version.

Regarding NETD…. Remember NETD of the SENSOR array is always specified with an F1.0 lens and not the lens fitted to the unit that is sold to the end user. That is the only way to ensure useful NETD comparisons. There is also the frame rate issue. The PICO640 specifications show 30fps associated with its NETD so the Pulsar product is likely providing a degraded performance due to slower optics and higher frame rate.

Fraser

[Fraser]

This page pretty much confirms to me that Pulsar are still using Lynred (was ULIS) microbolometers and we would most likely find the PICO640 core in the Accolade 2 product.

https://www.pulsar-nv.com/glo/zh/blog-news-577/thermion-2-xp50:26/

Pulsar are proud of their partnership with a well known and respected Microbolomter manufacturer.
Note that Pulsar clearly state that the microbolometer offers <25mk NETD and that they have fitted a f1.0 Germanium lens to it. This is superior to the Accolade lens system.

[Fraser]

Pulsar Thermion review may be of interest as it compares versions.

https://www.thefirearmblog.com/blog/2021/03/11/new-thermal-sight-pulsar-thermion-2-xp50-xq38-xq50/

[Fraser]

https://www.iwa.info/en/ausstellerprodukte/iwa21/exhibitor-76469362/pulsar

[Fraser]

With regard to claimed NETD there is also another factor to consider….

Microbolometers used to be temperature stabilised and this improved performance. The temperature stabilisation permitted the fine tuning of the microbolometer for maximum performance to create a ‘Hot’ (my terminology) core that performed better than cores that used the same parts, but were not optimised during configuration. Many microbolometers now operate in a non temperature stabilised mode where the microbolometer die warms up to a point of thermal equilibrium with its surroundings. The die temperature tends to reach thermal equilibrium at ~30C in a 20C ambient temperature environment. This uncontrolled die temperature means that specifications like NETD are often followed by ‘fine print’ to qualify them. In the case of the PICO640 core this ‘fine print’ includes a frame rate of 30fps and a temperature of 300 Kelvin. At other frame rates and temperatures, the specification is not guaranteed 

I broached this subject with FLIR some time ago in relation to their non temperature stabilised imaging cores. At first they were hesitant and coy in their response but I eventually got my answer. Basically non temperature stabilised cores only meet the stated specifications and accuracy figures at the temperature at which they were calibrated. Any deviation from that environmental temperature can degrade the performance, especially where temperature measurements are concerned. A high performance thermal imaging core may employ the classic Peltier temperature stabiliser on the microbolometer to guarantee its performance over a range of ambient temperatures and to enable the fine tuning of bias voltages for optimum performance. The price that is paid for temperature stabilisation is increased power consumption which can be an issue in mobile, battery powered equipment.

Fraser

[Fraser]

With regard to my above comments, upon re-reading them, I should clarify that most thermal imaging camera systems do meet their claimed measurement accuracy figures within a specified environmental temperature range. This is achieved using temperature sensors within the cameras casing that measure environmental temperature and the system applies measurement corrections based on calibration tables. The FFC flag, where present, also aids measurement accuracy as its temperature is used as a reference point when placed in front of the microbolometer.

My question to FLIR related to the increasing measurement error present in the imaging core as its environmental temperature moved away from the common 20C found in homes. The response was that the environmental temperature correction tables were the same across all cameras and so generic. The camera specific accuracy of those correction tables becomes worse the further away from the optimal environmental temperature the core goes.

[Max Planck]

Hi folks

I recently came across the not-cheap Pulsar Accolade 2 LRF Pro thermal monocular that claims a remarkable <25mK NETD from its 17µm 50Hz 640x480 uncooled LWIR sensor and 50mm f/1.2 germanium lens. Certainly, the few bits of video I've seen do suggest it's an excellent performer.

I can't afford the four to five thousand pounds it'd cost here but I wonder if anyone else has seen it, possibly even in the flesh? Any idea what sensor is in it?

And yes I do know that NETD figures can be massaged to the point of meaninglessness, even if there was a meaning there to begin with, but the pictures do look rather good, if heavily processed.

Feasible using VOx cores from SCD.

Max

[Fraser]

Max_Planck,

Pulsar have confirmed that they are using a new Lynred microbolometer.  It is A-Si and likely the PICO640 low NETD version.

[Bill W]

This page pretty much confirms to me that Pulsar are still using Lynred (was ULIS) microbolometers and we would most likely find the PICO640 core in the Accolade 2 product.

https://www.pulsar-nv.com/glo/zh/blog-news-577/thermion-2-xp50:26/

I find it somewhat odd that the Chinese page carries a claim that the 'sensor is from Lynred USA'
There does not seem to be a matching English or French page.

The sensor is undoubtedly made in France, I doubt an equivalent US made one would be freely exported to China.

[svgurus]

Pulsar discussion, hehe) Ultrapurple, its strange you didn't mention Helion XP2 pro, it has f/1 lens and still can take older wide/adapted ones, accolade has more problems with that afaik). There are 2 Pulsar assembly facilities that i usually hear of - in Russia and Lithuania, Lithuanian ones don't have warranty in Russia so usually i read reviews for Russian ones, and they mostly say new sensors are Chinese, maybe Lithuanian factory has access to Ulis/Lynred sensors) there are tests on youtube where they change lenses between 40mk xp2 and 25mk xp2pro https://www.youtube.com/watch?v=sRrYJB4QDek&ab_channel=KEENGAZE

and xp2rpo with older lens still looks more crisp, so i guess 25mk figure is close to real. I had a chance to get xp2pro at about 3200$, not sure about warranty but decided to wait, now they are sold out till autumn. The thing about it that bothers me is its non-removeable image upscaling, it has XGA output just like screen resolution, even for photos, and sensor is VGA. but video on youtube with double upscaling looks nicer from XP2pro than xp2))

25mK, 50FPS, but not  at the same time...

Iray upgraded their 12m sensors to 25mk too, recently, with 50fps of cause. but recorded video fps is still far from 50

BTW anyone knows rather easy test for real NETD? apart from counting noise on histogram of grey image

[Ultrapurple]

I find it somewhat odd that the Chinese page carries a claim that the 'sensor is from Lynred USA'
:
The sensor is undoubtedly made in France, I doubt an equivalent US made one would be freely exported to China.

Yes, that's what I meant - that it wouldn't be using a sensor made in USA and/or subject to their restraint of trade laws.

@svgurus - thanks; all very interesting stuff. Agreed that a non-scaled output would be useful...

Bottom line appears to be that the state of the art (at least in the open market) has advanced somewhat in recent years. If a manufacturer can reliably produce 25mK (presumably 17 or 15µm) sensors in quantity, I wonder what the cherry-picked best-of-the-best can do? But more importantly, the fact that there is so much more production today means more money for R&D and so we get that bit closer to my wish for a thermal equivalent to a DSLR - multi megapixel resolution, interchangeable and/or zoom lenses, non-bankrupting price.

[Ultrapurple]

Just to add to the above, the still well-regarded 12Mpix Nikon D700 has roughly 8.5µm square pixels on its full-frame 36x23.9mm sensor.  Yes, that's a big area for a lens to cover but a 50µm 320x240 sensor was 16x12mm and there were 'plenty' of f/1 optics for those.

Using the same maths, a 16x12mm sensor of 8µm pixels would be 2000x1500, or 3Mpix. I think that's roughly the state of the art for resolution now (if you've got the money).

[bap2703]

Just adding some late comments to the discussion.

Usually in the uncooled thermal imaging barely sensitive enough sensors world, the number of Hz you get is loosely related to how fast the sensor is.
That's because the readout electronics can be decorrelated to how fast the sensors responds to any change in the scene it sees.

Hence you can easily sell many Hz which are just a lot of supersampling of a slow sensor. (hello thermopile manufacturers)
So the value you attribute to the Hertz in the datasheet should be just that: the sampling rate.
If you buy hundreds of fps with a sensor that's sluggish, you will get a nicely sampled sluggish goo of pixels.

What you want to know is the response time of the sensor: how fast it senses a change in the scene.

The retail end user market being small and opaque, with no standardization like EMVA, it's especially easy to get crap 

[Bill W]

Just adding some late comments to the discussion.

Usually in the uncooled thermal imaging barely sensitive enough sensors world, the number of Hz you get is loosely related to how fast the sensor is.
That's because the readout electronics can be decorrelated to how fast the sensors responds to any change in the scene it sees.

Hence you can easily sell many Hz which are just a lot of supersampling of a slow sensor. (hello thermopile manufacturers)
So the value you attribute to the Hertz in the datasheet should be just that: the sampling rate.
If you buy hundreds of fps with a sensor that's sluggish, you will get a nicely sampled sluggish goo of pixels.

What you want to know is the response time of the sensor: how fast it senses a change in the scene.

The retail end user market being small and opaque, with no standardization like EMVA, it's especially easy to get crap 

Generally this is not the case.  Sensors are between 2ms and 10ms response time, 2ms being ASi and 10ms tending to be VOx.
Any of those could happily read a wholly 'new' scene at 60Hz.

Now whether the attached electronics can do all the necessary data shovelling and sums on all the pixels 60 times a second is a different question, or indeed if the designers even thought it worthwhile putting in the power and accepting the loss of sensitivity.

I'd suspect that a fair bit of your 'sluggish goo' is actually frame averaging to get rid of noise.  Additionally some of the 'export' FLIR 7.5 Hz cameras (K65 etc, so Tau cored) were particularly bad for that as they were averaged down to <9Hz.

Bill

[Ultrapurple]

<snip>
The retail end user market being small and opaque

I rather suspect that the retail market is becoming a good deal larger than the 'traditional' military (followed by 'professional' markets, eg firefighters, S&R etc). Nowadays one can buy a complete, functional thermal imager for about the same amount as a 'professional' user pays for a spare battery, or a high-end military user pays for a lens wipe. I don't see thermal imagers being given away in cereal packets anytime soon, but I do expect them to become more and more mainstream. Even today one can buy two or three pretty good 384x288 thermal cameras for the price of a high-end mobile phone...

... but you're right about the opacity.

[NCG]

Any known somewhat cheap commercial thermal viewers with the PICO1024(1024x768) 17µm? Pulsar might have killer product there...

[Ultrapurple]

Somewhat cheap commercial thermal viewers with the PICO1024(1024x768) 17µm?

WANT!!!

[svgurus]

Its funny someone sees Pulsar as kind Santa-Klaus)) Guess their lenses dont support this resolution and pixel pitch, making new lenses and making users buy them sounds not market-adequate. There is thin possibility that 1024*768 resolution in Pulsar Helion xp2pro with last firmware makes it more sharp than Iray 6's, but i must see it myself in autumn hopefully. I own Olympus cameras that make photos at 50mp handheld or from tripod while having 20mp sensor so maybe there is some good-working algorhythm that makes supersampling output at 1.6* resolution on each size, that will give you XGA without crazy Fluke/Flir prices

[Ultrapurple]

I didn't say Pulsar would/could be the source of an affordable XGA (or SXGA+) device, only that I would like something like that.

I'm already using superresolution to get 1280x960 from my 640x480 Therm-App Pro; the results please me. I can only imagine what a decent sensitivity SXGA sensor would be like with superresolution...

[svgurus]

Ok, I forgot to mention - Olympus makes astoundingly sharp superres)) there are a lot of praisings and reviews across the internets) even hand-held, though it takes 5-10sec to merge their 8-16 photos in-camera, but its 20mp source photos, again, not 0,3.
Yours are soft as regular photo through couple of window glass layers, sorry.
My point was that most LWIR superres gives 2x resolution on each side, i only saw thermviewer results for my therm-app and 301(they are lousy) and yours from therm-app pro on flickr(some were quite good). But Olympus makes 1,6x and 2x superres, 2x using sensor shift and 1,6x in 2 newest cams using hand instability afaik, and it gets work done great too, so maybe XGA output on newest Pulsar VGA thingies is not as lousy as I thought.

[Ultrapurple]

@svgurus With respect, when you say the thermal images are soft you are not comparing like with like. It's akin to expecting the images from an under-illuminated VGA webcam to compete with those from a quality DSLR.

Some quite high resolution thermal images have been made by combining superresolution with stitching. Although it takes time (and requires an appropriate lens), even a 320x240 thermal imager can make pictures of tens of megapixels. Those are the ones that may reasonably be compared with similar pixel-count visible light images.