When And How To Specify For Low Light Cameras

The cameras installed in any facility would have varying light intensities (lux) available.

My questions:

1. What is the lux level below which low light parameter should be specified?

2. What is the specification statement for the low light parameter.


Good question, tough application.

First important point: the lux levels you will get measuring light in the real world will not match / relate to lux levels specified at a camera. For example, you might be planning to install a camera at a parking lot with .5 lux measured (e.g., Training: Using a Lux Meter) but if you use a camera specified for .5 lux, it will almost certainly be completely blacked out, because camera manufacturers use super aggressive / unrealistic minimum illumination levels (see: Don't Trust Lux Ratings).

Second, a question: Can you not use cameras with integrated IR? There are definitely cases where integrated IR is not appropriate (e.g., wanting color images at night) but, for most uses, integrated IR is an easier and more reliable way to get decent low light images.

For your question 1, what do you mean by low light parameter? What type of low light parameter?

For your question 2, do you mean your specification? As in what should you specify? If so, that's really tough because vendors often lie about their minimum illumination spec, just because Vendor A lists .0001 lux and Vendor B lists .01 lux does not mean Vendor A will perform better (e.g., This Is Why You Cannot Trust Tri-Ed / Speco Super Low Light).

For example, you might be planning to install a camera at a parking lot with .5 lux measured (e.g., Training: Using a Lux Meter) but if you use a camera specified for .5 lux, it will almost certainly be completely blacked out, because camera manufacturers use super aggressive / unrealistic minimum illumination levels .

Not exactly...

Camera lux ratings are generally specified in terms of light entering the camera lens/sensor (this is a simplification, but accurate enough for the point of this comment).

Cameras see light reflected off objects. Unless your "object" is a mirror, only a small fraction of the light hitting the object is going to be reflected back off the object. Light also dissipates over distance (Inverse Square Law).

If you measure .5 lux in the parking lot as the general ambient light level, you'd have about .1 lux of light reflecting off a human subject (best case, as a rule of thumb clothing will reflect about 10-20% of the light hitting it). You'll also have some amount of dissipation, which depending on the distance may or may not be significant. Your .5 lux camera would be "seeing" a little less than .1 lux of light from the person you want to get on video, which as you noted will result in a mostly black image.

Lux ratings may sometimes be overstated, absolutely. But they are a measurement of light at the camera (which can be known/controlled/tested) not light from an unanticipated subject at an unanticipated distance (which can't be known or even realistically guessed at).

My personal rule of thumb for having hope of getting a decent image has been 2 Lux of light on-target per 100' of distance from the camera. EG: if you want a decent exposure of a person 200' away from the camera you'd want a minimum of 4 lux of ambient illumination at that 200' distance. This is by no means scientific, but it seems to work pretty well for most typical night time security camera applications.

"Camera lux ratings are generally specified in terms of light entering the camera lens/sensor (this is a simplification, but accurate enough for the point of this comment)."

Examples please.

From what I have seen most manufacturers provide no clarity about how it is measured but I frequently see manufacturers show lux levels with a lux meter display inside the box they are testing / demoing (like so).

Also, if you are right, how does Speco get a specification of 0.0005 lux?

From what I have seen most manufacturers provide no clarity about how it is measured

It's a double-edged sword. It *should* be done based on controlled tests, but most people won't understand what would be a "true" test (eg: a test of the sensor only, no lens) and how to work UP from there (if my raw sensor has a .1 lux min illumination specification to get a 100 IRE output, but I really only need a 50 IRE level and I'm using an f/1.4 lens, and my target is going to be 300' away, but I'm trying to get an exposure of white panel trucks, so they're pretty reflective, I need about 2.5 lux at my target).

The VideoIQ datasheets (Example) gave a sensitivity reading along with the f/stop of the lens used. There were (IIRC) based on 100 IRE output readings. They didn't specify shutter speed, but (also, IIRC) that was based on 1/30s or possibly 1/15s, but no slower.

This link is some good reading on the topic.

but I frequently see manufacturers show lux levels with a lux meter display inside the box they are testing / demoing (like so).

IMO that is not a scientific or specification-level test, but it's also closer to what most people would want or expect. Some kind of real-world reading, though most people still couldn't accurately extrapolate anything from that. The in-box test is also deceptive because far less light is scattered away from the scene, it's reflected around and contained inside the box.

My primary point is that the min lux ratings on a camera are never going to be something that you can use as an ambient light level specification for a subject that isn't immediately in front of the camera.

Also, if you are right, how does Speco get a specification of 0.0005 lux?

No comment.

"The VideoIQ datasheets (Example) gave a sensitivity reading along with the f/stop of the lens used. There were (IIRC) based on 100 IRE output readings. They didn't specify shutter speed, but (also, IIRC) that was based on 1/30s or possibly 1/15s, but no slower."

Brian,

How do you measure IRE for IP cameras?

Thanks

I am curious also, though one could always use a decoder, no?

Brian,

How do you measure IRE for IP cameras?

Thanks

In our case, we used the analog out of the sensor. This was very straight-forward for the D1 cameras, but became a little less direct with the 1080p cameras.

But there really is no direct way to measure IRE with an IP camera in terms of the encoded video signal. But in most cases you can get a composite output from the sensor assembly, though it may often require additional components that aren't part of the finished camera. This makes it hard (impossible) for someone verify or replicate the readings.

Overall I don't think it's worth getting hung up on the details of low-light readings. IMO from the IPVM tests most of the mainstream cameras are fairly equal within a given class. By the time you start really seeing the differences in low-light performance you are down to light levels that nobody would actually recommend using if you are concerned about low-light performance, and even the "best" performers tend to have fairly noisy video.

As new technologies come about (1/2" sensors, Light Catcher, etc.) it's worth re-evaluating what the "new low light" norm might be, but I still think in every case you'll find that moonlight alone is rarely (if ever) going to be sufficient for good images at night. You'll always need some amount of supplemental illumination, and once you're at the stage, the costs of adding 6 lux or 60 lux aren't all that different many times.

...in most cases you can get a composite output from the sensor assembly...

What format would the composite out for a 1080p sensor be in, NTSC?

If you measure .5 lux in the parking lot as the general ambient light level, you'd have about .1 lux of light reflecting off a human subject.

As I'm sure you know lux can be measured two ways, incident and reflective. If all you say is true then the OP should just be able to measure the reflected light in the parking lot from the camera POV, for a given subject, and then if that is greater than the mfr spec, the image won't be black.

Would you agree with that?

Would you agree with that?

Yes and no. Technically, yes, if there were some way to make that measurement you could do so. But practically speaking, holding a light meter next to the camera is going to give you a reading of all the general light in that area and also wouldn't account for any loss of the light through the camera lens.

There is no *practical* way to measure light at the camera placement point and get an idea of how a subject some distance away will appear on video.

You could have lots of light *at* the camera, and be trying to detect a person in a shadowy spot (underneath an unlit building overhang with lots of trees around), or you might have pitch black at the camera and a helpful burglar 200' away shining a maglight on his face. No at-camera reading is going to accurately measure either of those scenarios.

There is no *practical* way to measure light at the camera placement point and get an idea of how a subject some distance away will appear on video.

TTL metering?

Show me a practical example that could be used in the field.

Ok, you're on! ;)

Save your receipts...

My accountant told me I had to wrap everything up today to take the tax write off on business expenses so here goes:

Let's take an Avigilon 4K PRO camera, 35mm sensor, Canon glass;

It claims minimum illumination of 0.1 lux.

Then let's take a Canon(or compatible) SLR/DSLR 35mm sensor with TTL metering.

Point the camera at the subject and adjust the FOV to match the desired subject. Let's say for example this is a parking lot and we are interested in capturing cars coming thru a one lane choke point from some distance.

Frame a given subject using the DSLR, e.g. Gray mid-size car, during the low light period.

Set the metering mode to the whole FOV. Set the ISO to 100. Let the EV compensation adjust both aperture and shutter speed.

Using the cameras values calculate luminance using the standard photographic formula:

Luminance = 12.4 x fnumber^2/ (exposure time x ISO)

So for instance if the camera says f/2.8 and 1/30 sec that's ~160 lux reaching the sensor.

If the calculation gives you 1 lux, and the Avigilon spec is accurate, you will know that the Avigilon camera, with the same lens and FOV, is capable of detecting the car, at least in the best case scenario.

Honestly, it's just a theoretical stab in the dark, but you (possibly) have all the equipment, I would be interested in what the method yields vs the actual camera.

A) I hardly consider this a "practical" approach

B) Your DSLR is going to give you exposure data based on its own sensor, especially if you're allowing it to set aperture and shutter speed. You might very well end up with a set of readings based on values (1/5s exposure for example) that wouldn't be viable for a CCTV application.

However I do have the Avigilon camera, almost every lens Canon makes, and a 5D at my other house. I might try this experience next time I am there just to see if it yields anything usable. But even so, I don't consider this the kind of test/setup that you could reasonably expect someone to execute. It's far more complex than walking out to a spot with a light meter, taking a reading, and factoring my 2 lux per 100' of distance guideline.

Even if your camera is rated for .1 lux, you wouldn't want to *actually* give it only .1 lux of light, that's never going to result in the best image. You need to deliver significantly more light to the sensor to get what most customers would consider a worthy image (IMO). I don't think trying to solve for minimum illumination is ever really the right approach here.

A) I hardly consider this a "practical" approach

Practical is relative I suppose. It certainly could be "used in the field". I mean if it actually delivered anything useful.

B) Your DSLR is going to give you exposure data based on its own sensor, especially if you're allowing it to set aperture and shutter speed

Yes, however Canon, Nikon, Sony calibrate their sensors for auto exposure before they leave the factory, so theoretically any camera should yield equivalent EV's.

Allowing it to set aperture and shutter speed is the whole point. It's the way the camera 'tells' us how much light it is seeing.

You might very well end up with a set of readings based on values (1/5s exposure for example) that wouldn't be viable for a CCTV application.

Yes, the settings most likely will not be viable for CCTV. But we're not using them to set the Avigilon camera, only to determine the luminous flux reflected from the ROI. And then comparing that to Avigilon's stated claims.

Even if your camera is rated for .1 lux, you wouldn't want to *actually* give it only .1 lux of light, that's never going to result in the best image. You need to deliver significantly more light to the sensor to get what most customers would consider a worthy image (IMO). I don't think trying to solve for minimum illumination is ever really the right approach here.

Agreed again. For this reason I made my threshold 1 lux, not 0.1 lux, above. Fudge factor to taste.

I don't think trying to solve for minimum illumination is ever really the right approach here.

Do you think Avigilon's minimum lux spec has any utility at all? Maybe as a relative measure between other Avigilon cameras?

However I do have the Avigilon camera, almost every lens Canon makes...

Even this 'mother'? ;)

John. This seems like a good undertaking for you to arrive at a spec that can be fair to both the owner and the integrator. You've given us a long list of why this spec (and a number of others, including ppf) should not be relied upon and that's good info; but I would like to see an actual answer to this and similar issues. Rather than specifying a low light number, perhaps a performance procedure would be in order. Say taking a known light source (illuminating the target at specific levels), target and distance (easy to duplicate in an integrator's shop) and then requiring a specific result "category" (for lack of a better term). The Rotakin categories seem to be a good start (A-M). At some point however, only LL cameras or a supplemental IR would be needed. Here is the full test procedure from the UK

As a specifier i could then give, say 10 allowable manufacturers (with other specs), that i believe can meet this standard. And list which categories apply to each camera on my camera schedule. Maybe the manufacturers could eventually say which IPVM Standards they meet or exceed.

I know you don't want to be another UL, but someone should do it... and IPVM is the best. If not you who? Also, what do you think of the Rotakin Test procedure as detailed here.

We tried a similar type procedure here: Ranking IP Camera Low Light Performance. We focused on lines discernible but there's also a variance in how relatively bright or dark cameras are.

I believe Rotakin is hard coded for PAL / SD analog and I have not seen any significant adoption outside UK / Commonwealth for many years.

Yes I am aware of this test, but it involved specific cameras. If I am Specifying a generic camera in a bid situation, I have to start from scratch and even then I cannot be sure of the results (I, nor any of my colleagues, have the resources to test every camera), particularly if, as you so clearly indicate, we cannot rely on manufacturer's specs. If we had a performance spec with a reasonably understandable testing procedure (i.e. the Rotakin Test) perhaps it could be done on a sample basis or every camera for that matter. Or simply done on one camera of each type, no? Another possibility: have a test scenario with the needed outcomes specified for one or more "standard test situations" (i.e. an IP camera looking at a test target at xx feet from the image device, under xxx lux of lighting must yield a level X image on the Rotakin Test). The test can be done by the Owner's Representative after the bid and prior to awarding a contract to the lowest acceptable bid.

Its easy to close the bid on a specific camera simply by making the spec so tight others cannot bid. I would prefer to follow a practice that would make the spec job specific rather than manufacturer specific, so i can get the best balance between cost and performance needed for the job at hand.

Integrators out there, what do you think?

sounds like a performance spec to me. Drag out the resolution chart criteria.