POLL: What's the lowest resolution is useful for you?

[Logan]

It's too silent here recently, so I'll post a poll.
Numbers are approximate, eg. if you want to answer 1280x720, then just choose 1024x768.
This is the number that is bearable for you, not optimal, because I believe everyone want a 4k thermal 

[Fraser]

Resolution is very important but so is IFOV 

I owned a very nice Ganz VGA thermal CCTV camera that had such a large FOV lens fitted that it’s IFOV was no better than a QVGA camera with half the FOV. So what ? You might say, Well, where image detail is concerned, the combination of microbolometer resolution and optics FOV is important. This is why some of the 160 x 80 pixel cameras had narrow FOV optics fitted of say 12 Degrees instead of the common 24 Degrees or larger found on QVGA cameras.

Fraser

[Vipitis]

I have 80x60, 160x120, 384x288 and 640x480 cameras. And while interpolation on the first two really ruins the experience (as well as framerate). QVGA+ is adequate, to the point where VGA does give better results - but it's not by much. it probably down to the different pixel geometry and effective area as well as lenses and calibration. But I do really want to experience higher resolution for my intentions: photography and filmmaking.

[Bud]

I selected "Even lower" because there may be use cases where just a few by few pixels may be sufficient, such as pest deterring water sprinkler for the backyard.

[Fraser]

Bud,

An interesting take on sensor resolution 

The common PIR movement sensors are often just 2 or 4 pixels and very effective as a movement detection system sensor….. though not really a camera in the true sense of the word. I have looked with interest at the low resolution thermal imaging sensors from Panasonic and Melexis etc. They are capable of some decent if not very detailed,  thermal detection. That may be enough for some intelligent target detection or monitoring systems though. I remember IRISys producing a true thermal imaging camera, the IRI1000 series with just 16x16 pixels interpolated up to 128 x 128 pixels ! The same low resolution sensor array was used in their very successful people counting and tracking systems that are still used in retail and building management systems. Projects existed that used these 16 x 16 pixel people counters to identify adult, child, animal, male, female and walking route with breadcrumb trail ! It was all useful retail ‘intelligence’. In case anyone is wondering, the camera looks down on the targets and male/female detection was by walking gate rather than more obvious body shape difference. Yes Men and Women walk differently.

It has been interesting to watch the microbolometer sensor array market develop since I first used them in the FLIR PM570 in 1997. The sensor FPA’s have become so much smaller and cheaper. IRay are making some very nice FPA’s and InfiRay cameras that incorporate those FPA’s are performing well.

It will be interesting to see how the microbolometer market develops….. will be see manufacturers concentrating on higher and higher resolution FPA’s or will they focus on reducing the production cost of the common resolution FPA’s that we see in use today. Maybe we will see both, with the lower resolution FPA’s becoming more affordable and the higher resolution arrays made available at the current, relatively high,  prices of medium resolution (640 x480) FPA’s ?

[Bud]

...useful retail ‘intelligence’. In case anyone is wondering, the camera looks down on the targets and make/female detection was by walking gate at her than more obvious body shape difference. Yes Men and Women walk differently.

This is amazing. I now wonder if cats and possums walk differently, so the possum would get jet sprayed and my cat would not 

[Fraser]

You could attach a simple transponder to your cats collar that disables the automatic spray system while kitty cat is in the ‘danger zone’ 

[Fraser]

An area where we may see greater use of low resolution thermal imaging sensor arrays is in fire prevention and detection. Static Thermal imaging cameras are already used to monitor plant rooms for overheating equipment and there have been thermal cameras developed that look for the ‘tell tale’ shape and flicker frequency of a fire, even at quite long range. Obviously for longer range working higher resolution is needed. With the decrease in microbolometer FPA costs we may see early warning fire alarm sensors that can not only detect the thermal energy but also characterise its location and shape/size for an appropriate response. You could even have them monitoring the human occupancy of an area that is involved in a fire so that firemen know where to target their resources.

[Fraser]

This poll does not include the sub-QVGA resolution arrays that come between 160 x 120 and 320 x 240. These ‘slightly less than QVGA’ arrays are becoming very popular in consumer grade cameras and have much to offer. A common modern FPA size is the 256 x 192 pixel array but there are others that do not fit in the usual ‘pigeon holes’ of thermal FPA sizes. Guide offer the TIMO core with an FPA of 120 x 90 17um pixels.

[Ultrapurple]

I voted 320x240 although the 384x288 sensors are IMHO the 'sweet spot', at least at present, in terms of bang per buck. Although the numbers don't seem much different, the extra pixels really do make a huge difference.

I tend to use only 640x480 sensors at present, often with superresolution to 1280x960. VGA thermal is wonderful and I decided some time ago to stop buying QVGA-class cameras (a decision I have failed to live up to...).

I would love to get an uncooled LWIR camera of XGA or better, but I think it's going to be a while before such a desirable toy becomes even remotely affordable.

[Logan]

This poll does not include the sub-QVGA resolution arrays that come between 160 x 120 and 320 x 240.

Thank you for suggestion.
But it's impossible to include every resolution, so I just list the most common ones, people can choose the closest option if their answer is not in the list.

[Fraser]

Logan,

This was not a criticism, just highlighting that we are now seeing some different resolutions to the traditional ones that we know and love 

[CatalinaWOW]

I think Fraser hit the nail on the head with his comment about IFOV.  Probably the most common use of IR sensors is to identify objects whether it is a fire, a cat, a possum, a person or a vehicle.  If you discard special cases like fire where extreme temperature difference and unique time varying signature can provide single pixel identification it generally requires a significant number of pixels on the object to perform identification.  The numbers are not rigidly defined but generally it takes numbers like a few dozen to classify objects (trucks vs cars, cats vs people and the like) and numbers in the hundred range and up to do identification (my cat vs your cat, Ford vs Toyota and the like).  Classification can be achieved with fairly small arrays, with the 16 x 16 array that Fraser mentioned near the bottom end.  But obviously very good camera pointing is required to achieve the required on object counts with a very small array.  Either larger fields or scanning is required to locate such objects.  Identification requires a larger minimum array size on the order of 30 by 30.

It seems that for human comfort several of these fields across the image is about the minimum, which is why analog TV had delivered resolutions on the order of 300 to 500 pixels on a side.

So I would give a tripart answer.  About 16 x16 is the minimum needed while desire starts to be satisfied a little before VGA resolution.  But if cost and size were no object I wouldn't stop before you got to astronomy size arrays, several thousand on a side.  Once an image is captured you can spend a lot more time examining it in detail.

[Bill W]

Probably the most common use of IR sensors is to identify objects whether it is a fire, a cat, a possum, a person or a vehicle.  If you discard special cases like fire where extreme temperature difference and unique time varying signature can provide single pixel identification it generally requires a significant number of pixels on the object to perform identification. 

FLIR articles cover this quite well, using the 'Johnson criteria' for detection, recognition and identification:
http://www.flirmedia.com/MMC/CVS/Tech_Notes/TN_0002_EN.pdf

and Opgal have a generic calulator
https://www.ophiropt.com/infrared/calculator/dri-range-calculator/

The original work of Johnson though was done for visible 'targets' in night vision systems

The flaw as you noted is that a distant light (ie active source) is going to be detectable by a single pixel and quite likely even if it is below one pixel field of view.  Now in thermal of course ALL targets are active sources and once a single pixel is lifted by a few times the MDTD you will see it, you will not need to fill the classical 2 pixels.

Conversely there are atomospheric effects acting the other way, usually ignored.

A much longer read...
https://www.ida.org/-/media/feature/publications/a/at/a-tutorial-on-e-lectro--opticalinfrared-eoir-theory-and-systems/ida-document-d-4642.ashx


Bill

[IwuzBornanerd]

This is amazing. I now wonder if cats and possums walk differently, so the possum would get jet sprayed and my cat would not 

Examples attached.   You will need to download the files & change the mp3 to mp4 to see the videos, since there are no video formats accepted.

As for the survey,  I voted for the 160x120 since it is closest to the Seek 206x156 which I use for surveillance.  The attached files, though, are from the Seek Slow Frame Pro at 320x240.

[tszaboo]

I believe lower resolution is acceptable, if we can blend with real pictures. I used the seek thermal for example, and it was disappointing compared to the Flir cameras.

[Fraser]

tszaboo,

You raise a good point. The resolution of the microbolometer and resultant IFOV can effect the users interpretation of a scene. With purely thermal imagery I have found 160 x 120 pixels to be about as low as I would go where sense needs to be made of scene content. At 320 x 240 pixels the scene is more easily interpreted and the context of the ‘hot spots’ is easier to see. In many thermography applications the user is either searching for anomalies or wishing to identify areas of interest through measurement of thermal energy being emitted by them. It is relatively rare to need a pin sharp very high resolution thermal image of a target in professional thermography. By its very nature, thermal emissions tend to be somewhat vague when it comes to their boundaries as substrate conduction often blurs the edges. That said, it is possible to obtain very well defined thermal images if such is required and appropriate equipment used. The recent increase in thermal imaging being used for broadcast applications such as spotting wildlife, has lead to a desire by many to achieve similar performance from consumer grade thermal cameras. Sadly the cameras used by most broadcast companies for wildlife observation are both high resolution and very expensive. That is not to say that good results cannot be obtained with a 640 x 480 pixel thermal camera that is equipped with a suitable lens though. This is a different usage scenario to that normally seen in industry and domestic thermal camera applications and sadly it is likely to be an application of thermal imaging technology that requires cutting edge FPA performance and resolution. Why ? Because this is a move from utility use of a thermal camera to artistic use where the image quality and clarity is all important and the Radiometric measurement accuracy less so. “Pretty pictures”, as I choose to call them, can be very impressive but sadly the audience can be very demanding and critical as many are used to modern high resolution VL digital images. This increases people’s expectation of thermal image quality beyond what can reasonably be provided at this point in time by most photographers working in the thermal domain. It is a bit like producing the pictures for Vogue magazine using a cheap Logitech QVGA or VGA web cam from the 1990’s !

To enable users to make sense of what they are looking at and to provide context to the thermal image, some manufacturers have added thermal fusion and edge detection overlays to their cameras. The visible light information overlayed or fused with the thermal image can greatly increase the user friendliness of a low thermal resolution camera. FLIR knew this and introduced the MSX system in an effort to make their lower resolution cameras appear to provide more detailed thermal imagery than was actually possible at the provided resolution.
With the FLIR ONE G2 and G3 dongle cameras FLIR did not provide the option to switch off the MSX visible light overlay. I was told by FLIR that the marketing team wanted the images produced by these dongle cameras to appear higher resolution than they were in case people compared them to the higher resolution Seek Thermal cameras with which they were competing. The image fusion technology employed by FLUKE and others is a mixed blessing. Whilst it adds a colour visible light image overlay to the thermal image that can provide context to the complete image, the mixing of a colour VL image with a colour Thermal image can lead to a confusing mash up of colours in the scene. The use of a monochrome VL image and/or adjustable VL image transparency can assist the user in determining what is thermal data and what is VL data in a scene. The Achilles Heel of visible light image use on a thermal camera is its need for adequate illumination. Without adequate light levels in the region of interest, the VL camera struggles and a very noisy image can result, or even no image at all. This is why some cameras integrate an LED illuminator into their camera designs. The range of such illuminators is often quite limited however.

[Bud]

This is amazing. I now wonder if cats and possums walk differently, so the possum would get jet sprayed and my cat would not 

Examples attached.   You will need to download the files & change the mp3 to mp4 to see the videos, since there are no video formats accepted.

As for the survey,  I voted for the 160x120 since it is closest to the Seek 206x156 which I use for surveillance.  The attached files, though, are from the Seek Slow Frame Pro at 320x240.

Ha-ha, thanks! They do walk differently, the cat's walk was so elegant