How close does a camera need to be to identify people in photos?
May 10, 2017 3:38 PM Subscribe
Is there a rule of thumb or formula by which I could figure out whether a camera with focal length F and resolution R at distance D would capture sufficient detail to identify people in images?
I help organize an event where there's a strong expectation of privacy. There is also some photography, and this year, there also will be drones. I would like to be able to tell drone operators "you should stay above altitude X in order to be on the right side of the photography policy" if possible. I realize that shooting overhead will make people less identifiable, but not all shots will be top-down.
I help organize an event where there's a strong expectation of privacy. There is also some photography, and this year, there also will be drones. I would like to be able to tell drone operators "you should stay above altitude X in order to be on the right side of the photography policy" if possible. I realize that shooting overhead will make people less identifiable, but not all shots will be top-down.
I work in risk management. "There's a strong expectation of privacy" and "there will be drones shooting photos" isn't compatible, short of (and maybe even with) a strong and clearly-defined agreement with the drone photographers about what is and isn't acceptable in terms of sharable photos.
I am not your risk manager, but I would strongly encourage you to be sure that whatever participation agreement participants are asked to sign includes an acknowledgement that there may be photos taken and they may be identifiable in them. If that's not possible (whether because it would make a significant portion of your attendees wave off or for another reason) then you need to not allow drone photography. You probably also need to review your agreements with the non-drone-photographers regarding privacy, for what it's worth. If I were doing risk management for an event, hearing "there's a strong expectation of privacy" and "there is photography" would definitely make me hit the "ON HOLD until we figure out how this works" button.
posted by Lexica at 4:52 PM on May 10, 2017 [6 favorites]
I am not your risk manager, but I would strongly encourage you to be sure that whatever participation agreement participants are asked to sign includes an acknowledgement that there may be photos taken and they may be identifiable in them. If that's not possible (whether because it would make a significant portion of your attendees wave off or for another reason) then you need to not allow drone photography. You probably also need to review your agreements with the non-drone-photographers regarding privacy, for what it's worth. If I were doing risk management for an event, hearing "there's a strong expectation of privacy" and "there is photography" would definitely make me hit the "ON HOLD until we figure out how this works" button.
posted by Lexica at 4:52 PM on May 10, 2017 [6 favorites]
Axis has a nice guide intended to assist their customers with placement of security cameras such that they can be used to reliably recognize or identity individuals (they recommend at least 1.25 px - 5px/cm). NIST also has a more technical presentation in which they recommend 5px/cm to 10px/cm photography.
Do you and your attendees trust your drone operators implicitly? If so, then your goal should instead be to just reduce the amount of footage you need to review in order to prevent inadvertent disclosure. A good rule of thumb would be to use drones with standard HD video resolution or worse, place two flags 100' apart on the ground, and the altitude at which both flags are visible in a video frame from the drone camera is a good minimum altitude for your drones. Doing this will produce video footage that will require only minimal review.
On the other hand, if you and your attendees do not trust your drone operators implicitly, what is your threat model? If you don't trust your drone operators to review their footage for inadvertent disclosure, why would you trust them to accurately know and report the capabilities of their drones? I have no doubt that for less than $5,000 one could obtain a drone that would be able to take photography sufficient to identify an attendee looking upward all the way up to maximum legal ceiling for operating a drone (500'). Do you anticipate that someone might be willing to spend a few thousand dollars to violate your attendees strong expectation of privacy?
posted by RichardP at 5:36 PM on May 10, 2017 [3 favorites]
Do you and your attendees trust your drone operators implicitly? If so, then your goal should instead be to just reduce the amount of footage you need to review in order to prevent inadvertent disclosure. A good rule of thumb would be to use drones with standard HD video resolution or worse, place two flags 100' apart on the ground, and the altitude at which both flags are visible in a video frame from the drone camera is a good minimum altitude for your drones. Doing this will produce video footage that will require only minimal review.
On the other hand, if you and your attendees do not trust your drone operators implicitly, what is your threat model? If you don't trust your drone operators to review their footage for inadvertent disclosure, why would you trust them to accurately know and report the capabilities of their drones? I have no doubt that for less than $5,000 one could obtain a drone that would be able to take photography sufficient to identify an attendee looking upward all the way up to maximum legal ceiling for operating a drone (500'). Do you anticipate that someone might be willing to spend a few thousand dollars to violate your attendees strong expectation of privacy?
posted by RichardP at 5:36 PM on May 10, 2017 [3 favorites]
Depending on the demographics of your attendees, there may not be any useful {not closer than X] metric. If your attendees are largely white and able-bodied, for example, then dark-skinned people of color or people with visible disabilities are going to be recognizable if only by process of elimination.
posted by rtha at 9:08 PM on May 10, 2017 [1 favorite]
posted by rtha at 9:08 PM on May 10, 2017 [1 favorite]
It's 2017. You cannot have a 'strong expectation of privacy' in any place with cameras outside your personal control.
Cool drone shots are utterly incompatible with privacy. My old-tech Parrot BeBop will happily identify a person from 100m. It has a 20° off-axis capability. We can document your nostril-hairs while appearing to look in another direction.
Current drone tech is vastly better.
If your drone operators are part of your community, and they share your need for strong privacy, and their systems are entirely secure, and they are happy to delete all thir data after handing it to you, and you have experts in de-identifying video, then you *might* be able to bring the risk of identification to an acceptable level.
But drones and privacy are not compatible in 2017.
posted by Combat Wombat at 5:28 AM on May 11, 2017 [2 favorites]
Cool drone shots are utterly incompatible with privacy. My old-tech Parrot BeBop will happily identify a person from 100m. It has a 20° off-axis capability. We can document your nostril-hairs while appearing to look in another direction.
Current drone tech is vastly better.
If your drone operators are part of your community, and they share your need for strong privacy, and their systems are entirely secure, and they are happy to delete all thir data after handing it to you, and you have experts in de-identifying video, then you *might* be able to bring the risk of identification to an acceptable level.
But drones and privacy are not compatible in 2017.
posted by Combat Wombat at 5:28 AM on May 11, 2017 [2 favorites]
Response by poster: I am reasonably comfortable managing the policy side of this. When I have questions about that, I'll ask them.
I am interested in the technical aspects of resolution/identifiability as a complement to that.
posted by adamrice at 7:00 AM on May 11, 2017
I am interested in the technical aspects of resolution/identifiability as a complement to that.
posted by adamrice at 7:00 AM on May 11, 2017
Is there a rule of thumb or formula by which I could figure out whether a camera with focal length F and resolution R at distance D would capture sufficient detail to identify people in images?
No. There's also aperture, angle, light, quality of the sensor, quality of the lens, and other factors to make this practical unless you're going to be so extremely limited (drones must be 500 feet overhead) as to reduce people to a few pixels.
I've not played that much with facial recognition in the past few years, but I can tell you just the free stuff bundled into Picasa in 2009 or so was almost frighteningly good at finding and identifying people in the shadows in the background of shots.
posted by Candleman at 8:16 AM on May 11, 2017 [2 favorites]
No. There's also aperture, angle, light, quality of the sensor, quality of the lens, and other factors to make this practical unless you're going to be so extremely limited (drones must be 500 feet overhead) as to reduce people to a few pixels.
I've not played that much with facial recognition in the past few years, but I can tell you just the free stuff bundled into Picasa in 2009 or so was almost frighteningly good at finding and identifying people in the shadows in the background of shots.
posted by Candleman at 8:16 AM on May 11, 2017 [2 favorites]
This thread is closed to new comments.
posted by bitdamaged at 4:15 PM on May 10, 2017