Subscriber Discussion
View A Face 3000 Meters Away
Can anyone recommend a fixed/ptz camera that can cover a distance of up to 3000 m ,to recognize a person's face at this distance. I don't think any of the available gigapixel cameras will fit this budget. It is for installation at an airfield.
Wow, 3000m is really really really hard. Here's the basic math from a lens calculator:
Speeddomes are definitely out. Even a high end 40x SD PTX or 20x HD PTZ is unlikely to capture faces at more than 100m (see our How Far Can A PTZ See? guide).
Your best bet is to use a pan tilt positioner with a super long varifiocal lens. I am no expert in such lenses but here's one example from Fuji (that costs like $50,000).

For non metric-speakers: 3000 meters = 1.86 miles
wow!

IPVMU Certified | 01/25/13 04:42pm
I'd also think anything having to look that far out would make for a pretty unstable picture as any little vibration would cause shake. You'd probably have to have a floating platform for mounting. We're humans, we can probably do it, but won't be cheap.
I'm glad nobody has brought up day / night use, nevermind all weather. Maybe VUMII could rise to the challenge...

01/25/13 11:28pm
I'm not saying this is an unsolvable problem, but it IS going to be one with a very expensive solution. Can you clarify if this is for a fixed location, or if some kind of PTZ control is required? Do you need 24/7 coverage of the location, or only daytime coverage? What is the area like in terms of lighting/contrast/etc.? What kind of weather or atmospheric interference is common in the area? What are the end-user expectations (eg: it will work 100% of the time, it will work in ideal conditions, etc.).
The lens math on this is fairly easy. The solution is more complex.
I suggest you step back and redefine your project. Do you really need to see the face at 3000m? Would the presence of an individual in an unauthorized area be sufficient? You could dispatch security to determine who the person really is. Putting cameras at main entrance points could provide facial images.
You might find some insight from SightLogix.

01/26/13 01:13am
Wow, Aaron, that's quite a leap. The tightest shot you'd get would be a 4.6 degree lens, resulting in an HFOV of about 750 feet wide. FAR beyond anything that Sightlogix (or any analytics system for that matter) is going to have the slightest shot at detecting.
You've gone from "face recognition" to about .4 ppf in your recommendation there. You're correct in that they might not *really* need to see a full face, but they do still probably at least need to know it's an actual person...
I think Vummii is worth a look, hat tip to Skip for bringing them up.
Their best model for super long range imaging appears to be Discoverii, specifically the 3000 model.
Here's what they claim for distances at night in sheer darkness:
And here's a video example:
While it's not delivering faces at 3000 meters, it's pretty fantastically long, considering the pitch blackness it is doing it in.
You MUST check out this incredibly funny discussion on the LinkedIn CCTV group. Vivek asked the same question but the recommendations are hysterical, including:
- An SHDSL transmission system
- A wireless IP kit
- "if you want face or license plates at a mile use the Avigilon 29MP fixed iris camera with 5-5000mm lens"
Here's a classic from another forum - 16MP Avigilon camera with Canon EF-mount Sigma 300-800mm f/5.6 lens (list at about $8000) and 2X teleconverter for an effective length of 600-1600mm at f/11:
Won't quite get you facial recognition, even at half the distance, but it's sure a good start :)
Matt, can you remotely zoom that lens or is it manual?
Also, it's an f/11 lens? What will you see at night?
Recall that the Avigilon Pro cameras use the Canon EOS lens system - being designed primarily for use on a DSLR, the lens would not have power zoom, but it would have autofocus (I don't know if the Avigilon camera takes advantage of this - the lens has the AF motors, but the AF sensing and control itself is normally sourced in the camera).
The lens is f/5.6, according to the spec; the 2x converter makes it f/11.
You'll recall our thread about f-stops being simple math: the max aperture stays the same, but when you increase the focal length, the f-stop changes accordingly - two stops for a 2x, one stop with a 1.4x.
To wit: at 800mm and f/5.6, the aperature will have a diameter of 800/5.6, or about 142.86mm, or inversely, 800/142.86 = 5.6 (plus a bit for rounding errors). Double the focal length, but the aperture stays the same, and you get 1600/142.86 = 11(ish).
While that won't allow a lot of light, keep in mind that these cameras also have a full-frame 35mm sensor, which can collect a lot more light than your average 1/3" or 1/2" CCTV sensor, so what light you DO get, will provide a far cleaner signal off the chip, and allow for more amplification.
There are power zoom lens options:
Fujifilm D60x12.5R3DE-V41
12.5 mm - 750 mm (60x), 25 mm - 1500 mm (60x) with 2x extender (provided)
F3.8 - T1500 (Equivalent to F1500) F7.6 - T3000 (Equivalent to F3000) with 2x extender (provided)
"C" mount, Format: 1/2"
Focal Length: 10 - 1000mm / 20 - 2000mm (2EXT)
Iris Range: F4.0 - 256 / F8.0 - 512(2EXT)
Format: 1/2"
Mount: C
Bosch has a long range imaging systems called the GVS1000. It claims to "deliver detailed images up to 1 kilometer / 3200 feet away, so you can see faces, read markings, and determine the intent of potential intruders." That's about 1/3rd of the requested distance but fairly long range for an actual surveillance product.
Matt, thanks for sharing those lens - very interesting to see the physical difference.
As for comparing to short range surveillance lenses, my point wasn't about distance but about expectations of light needed. If someone has experience with F1.2 or F2.0 lenses, if they go to an F/11 they will understimate how much radically more light they need, even if it's a better camera with a bigger lens.

01/29/13 09:04pm
Regarding the LinkedIn thread, that's why I barely look at any of the other CCTV groups/discussions on LI.
Any basic question gets pummeled with useless replies "Dear sir I suggest you check out our product Bob Camera.". And then the suggested product fits between 0 and -4 of the key criteria that the poster was aksing about, or even better has all but non-existant online info.
The only one of Fujifilm which comes closest to it, is the HC16X100R2CE-F11
It has a focal length of 200 - 3200 using it's 2x extender.
Although it's only applicable from 1" to 1/2".
Advantage is that it's actually made for HD (3MP) camera's.
PS: It weighs 24kg!!!!
In the "can it even be done" category with money no object, how about an L3com WESCAM MX-20?
At 10,000 ft (a little over 3,000 m) it's marginal on faces -- absolutely not recognition quality. By 2,500 ft, you might be able to recognize a person. At over $1M/copy for airborne variants with dual visible and infrared HD imagers, it might be one extreme to bound the problem.
07/10/13 02:33am
What about Sony XIS?
Here's our review of Sony XIS. There's some pretty amazing elements of the system but it's not designed to capture faces thousands of feet away. Indeed, it's more of the opposite, monitoring very wide areas vs super far away locations.
07/10/13 02:48am
Good info. Thanks!
This camera came on the market a few months after the OP and does the job on paper:*
CohuHD 8800
Here's the breakdown:
Required Distance: 3000m
Focal length: 1680mm
H Sensor Size: 9.398mm (1/2.7")
HFOV = Distance / (Focal Length / H Sesnsor Size)
16.75m (55 ft) = 3000m / (1680mm / 9.398mm)
PPF = H Resolution / HFOV
35 PPF = 1920 px / 55 ft
Approx. Street Price: $35,000
*Admittedly 35 Pixels Per Foot is at the very bottom range for facial recognition.
I think the correct H Sensor size is actually 5.37mm. I noticed a month ago that the number was too large, but didn't say anything then because you were at the 'bottom range' and I didn't want to make it worse. But then I realized the smaller the sensor (assuming the # of pixels is constant), the greater the PPF, so now ~60 PPF
Required Distance: 3000m Focal length: 1680mm
H Sensor Size: 5.37mm (actual 1/2.7" sensor format width)
HFOV = Distance / (Focal Length / H Sesnsor Size)
9.58m (31.4 ft) = 3000m / (1680mm / 5.37mm) PPF = H Resolution / HFOV
61 PPF = 1920 px / 31.4 ft
Newest Discussions
Discussion | Posts | Latest |
---|---|---|
Started by
Georgi P
|
6
|
about 3 hours by Undisclosed #3 |
Started by
John Honovich
|
4
|
about 4 hours by Undisclosed #2 |
Started by
Brian Rhodes
|
1
|
about 7 hours by Brian Rhodes |
Started by
John Honovich
|
7
|
8 minutes by John Honovich |
Started by
Rick Caruthers
|
25
|
less than a minute by Dan Hoskins |