I am wondering if it is at all possible to have an IR camera that can be faithfully used in broad daylight. I have already seen this question here, but I am looking to understand if a full blown camera can even have this capability.

What I would like, is to be able to make out cars (that are on), and humans, against a background using an IR camera in broad daylight. Is that even possible? I would think so, but I am not sure of the physics/electronics constraints in this regard. Thank you.

  • 1
    \$\begingroup\$ Daylight includes IR, very sensitive cameras you have to filter out one or the other or the image will appear blurry. \$\endgroup\$ – John U Jun 19 '14 at 14:23
  • 2
    \$\begingroup\$ Near IR or far IR? If you want far IR ("thermal imaging"), the ambient heating of objects and air by the sun will present a problem. The overall intensity can be managed by decreasing the aperture just as with a visual light camera. \$\endgroup\$ – pjc50 Jun 19 '14 at 14:27
  • \$\begingroup\$ @JohnU Right, daylight does include IR, so I am thinking, do temperatures of humans stand out with good SNR relative to the ambient temperature of passive objects in broad daylight? \$\endgroup\$ – TheGrapeBeyond Jun 19 '14 at 14:30
  • \$\begingroup\$ @pjc50 I was not aware of near or far IR, I will look into that. But basically I think my application will be 'thermal imaging' as you said. That is, take an IR picture from say, a roof of a building, and from that picture, clearly be able to make out humans/cars relative to ambient background heat in daylight. (I should add that I would never be looking skyward, FWIW, only downwards). Thanks. \$\endgroup\$ – TheGrapeBeyond Jun 19 '14 at 14:32
  • \$\begingroup\$ Have a look at the example videos on FLIR's website, their cameras get used in daylight. \$\endgroup\$ – John U Jun 19 '14 at 17:37

It sounds like you are talking about Thermal Infrared, 10um - 14 um wavelength which corresponds to a black body temperature of 300 K. This is what is typically used to image and detect the body heat of mammals. The detector here are either MCT (Mercury Cadmium Telluride) cooled detectors or Bolometer based.

The answer is yes, the ambient visible light levels only affect the image in so much as there might be differential heating of exposed surfaces and those will appear hotter (which appears brighter).

In military systems for thermal targeting they use InGaAs detectors (3um - 5um) which corresponds to higher temperatures from exhaust plumes etc. But again operation in daylight is not an issue.

  • \$\begingroup\$ Thank you for your answer placeholder. I have some followups: Do you mean that the MCT / Bolometer based ones are not sensitive to ambient heat from the background, and so will give good SNR for body heat on a daylight background? I guess I am confused why you mention 'visible' light when we really just care about the IR spectrum.... 2) For the InGaAs, are you saying those can also be used to see body heat in ambient daylight background with good fidelity? Thanks! \$\endgroup\$ – TheGrapeBeyond Jun 19 '14 at 14:44
  • \$\begingroup\$ No I'm saying that the detector detects heat emitted by surfaces in that range of temperatures. Rarely is the background going to be be at 0K so there will be ambient background and differential heating which can confuse the image. the key to discrimination is the thermal discrimination ability called NETD (Noise Equivalent Temperature Difference). A low NETD will allow you to discriminate objects against a background of similar temperature. \$\endgroup\$ – placeholder Jun 19 '14 at 15:55
  • \$\begingroup\$ You are saying that for the most part we can use an IR camera to faithfully discriminate between hot human bodies VS a background during daytime, but that a source of confusion might be 'differential heating'? \$\endgroup\$ – TheGrapeBeyond Jun 19 '14 at 16:04
  • \$\begingroup\$ You're too fixated on the sunlight thing. The sensor only is affected by the sunlight (unless you're staring at the sun or getting glints) by the fact that things in the scene will also be at the same temperature as what you need to see. It's hard to see a white object on a white back ground in the visible. In thermal it's hard to see a mammal (at 300 K) against a 300 K background. At some point the differences are swamped by the noise of the sensor, a more sensitive sensor can discriminate better because these differences aren't hidden in noise. \$\endgroup\$ – placeholder Jun 19 '14 at 17:30
  • \$\begingroup\$ I get that the SNR is a factor. My question is, is ambient background in broad day light at, or hear, 300k? If it is, then we will not be able to distinguish a human. \$\endgroup\$ – TheGrapeBeyond Jun 19 '14 at 18:26

Yes, it is typically possible to use a thermal imaging camera in this scenario. The reason is that such a camera is fundamentally sensitive to contrast in radiant flux, which depends not only on temperature, but also emissivity. Since emissivity varies between different substances, there can be sufficient contrast in apparent temperature to distinguish even between objects that are at the same actual temperature.

For example, given a piece of bare metal and a piece of glass at the same temperature, the radiant flux observed from the (low reflectivity, high emissivity) glass will principally be due to its own temperature, while that seen from the (high reflectivity, low emissivity) metal will predominantly be reflected from surrounding objects. Depending on what these objects are, the metal may be distinguishable from the immediate environment (e.g. if it is reflecting the sky, which has an effective temperature of roughly -50°C), or it may not be. Telling high-emissivity substances apart when they are close in temperature is often more difficult. By the same token, the actual temperature of high-emissivity substances is influenced more strongly by radiation from their surroundings than that of low-emissivity ones, because the former absorb more of the incident radiation.

Note that I chose glass as an example of a high-emissivity substance to highlight the difference between our ordinary experience and the situation in the long-wave infrared (LWIR). Glass has low emissivity in the visible region, but appears very different when observed in the thermal infrared--it appears "black" (strictly, as a black-body) rather than transparent. Substances such as silicon show somewhat opposite behavior, with high emissivity in the visible, but low in the LWIR.

To distinguish differences in real temperature from differences in emissivity, the approach is to measure the radiant flux at two or more different wavelengths. Then, the temperature is recovered from the ratio(s) of the measured flux(es) to one another, rather than their absolute values. However, this is rarely done in thermal imaging since it adds cost and complexity to what is already a costly and complex device. More typically a radiometric calibration is done, but this is not robust to changes in the scene being observed.

The characteristics of the detector also dictate what is or is not a resolvable difference in radiant flux.

HgCdTe (MCT) detectors operate as photodiodes, so they have good sensitivity (noise-equivalent temperature difference, NETD, down to about 10-20mK depending on operating temperature) and fast (sub-microsecond) response. If the read-out circuitry is saturated through observing a hot object, this can be addressed by reducing the integration time. However, these detectors must be cooled, otherwise internally thermally generated carriers dominate. If detection is done in an a.c. mode using a chopped signal, the operating temperature can be quite high (-30°C). If it is operated in a staring mode, cryogenic temperatures (77K/-196°C) are typically required. Temperature variations in the optics and chassis also affect the imagery, which can be a problem for outdoor operation. In addition to the cooling requirement, HgCdTe focal plane arrays (FPAs) are very difficult to manufacture and typically exhibit extreme pixel-to-pixel response nonuniformity as well as many dead pixels. Hence, they require periodic non-uniformity correction using black-bodies at two different temperatures. After this is done, the images look very nice.

A typical HgCdTe FPA size may be up to about 320×240 pixels, but not more than that unless you are willing to spend extravagant sums of money for a product produced using a low yielding process. In fact, unless you are contracting for the US military, large HgCdTe detectors are probably not available at any price. The cost for the commercially available 320×240 detector may be roughly $50-100k. InSb FPAs are also available, and they have most of the benefits of the HgCdTe technology while avoiding most of its problems. However, because they are sensitive to shorter wavelengths, they are limited to temperatures above 0°C. And the cost is still high, of course, but not shockingly so.

The other common detector type used for thermal imaging is the microbolometer. These sensors use materials with a high temperature coefficient of resistance such as amorphous silicon (relatively easy to manufacture) or vanadium(V) oxide (higher TCR), and the FPA is constructed from a grid of tiny thermally isolated resistors exposed to incident thermal radiation. Because they measure temperature directly, they are sensitive over a wide spectral range. These detectors have fewer problems with uniformity, and because a change in resistance is being measured instead of a current, there is no problem with thermally generated carriers. The sensor can thus be operated at room temperature after initial calibration for the resistance of the individual elements. The readout circuitry is also cheaper because it can be row-column multiplexed, which is acceptable because the thermal time constant is such that instantaneous readout would not be possible anyway. However, it is difficult to manufacture the FPA so that the resistors are all equally well thermally isolated from the substrate and one another, and measuring their resistance does of course entail passing current through them, which can change their temperature. For these reasons the image does look more grainy than from a HgCdTe or InSb FPA, the NETD is higher, and the frame rate is lower. But, it is more than sufficient for many thermography applications, and the cost is much lower than the photoconductive detectors. With the rise of silicon-based microbolometers, they have essentially become commodity products.

If you are seriously thinking about thermal imaging for a particular application, the most appropriate advice would probably be to contact a manufacturer of thermal imaging cameras and ask for a demonstration. There are many companies serving this market, but FLIR Systems has an extraordinarily wide range of products and in recent years has been increasingly catering to scientific users, so they would be one of my first contacts personally.

Footnote: to clear up some terminology arising in comments: thermal infrared (roughly 3-20µm or sub-bands within that) is in the mid-infrared spectral region. Within this we have the mid-wave (3-7µm) and long-wave (7-20µm) infrared bands. These definitions are quite approximate and the borders between them depend largely on the availability of detector materials and the specific application. However, "far infrared" refers to wavelengths of roughly 20-100µm, which correspond to lower-than-ambient temperatures: this spectral region has few practical applications and is mainly of scientific interest. There are also the short-wave infrared (1.5-3µm) and near-infrared (700-1500nm) regions, which are sometimes used for non-contact temperature measurements of hot objects, but otherwise are not normally considered thermal.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.