I was sent this video where a man demonstrate his HP webcam is unable to facetrack him because of his skin color, He’s black. However, as soon as his white co-worker comes in, the face tracking works (http://www.youtube.com/watch?v=t4DT3tQqgRM). Consumer Reports did a response that apparently showed that it was a lighting problem, but it took some effort to get it to work (http://www.youtube.com/watch?v=NphmOV0lrBg). So, how does face tracking work and why would this problem happen? If it was the lighting, why doesn’t it focus on the light reflecting off his forehead?
My understanding of the problem (I’m a former pro videographer) is that it is primarily a lighting issue - video cameras, especially cheap ones like webcams, do not have a great contrast range. Think of your typical computer room set up - most rooms are painted white or some other light color, and you don’t generally have lights directed in your face when you’re sitting at the computer.
So, if you’re black, your webcam is looking at a very dark, unlit face contrasted with a very brightly lit background, creating loads of contrast. The face is the smaller element in frame, so the camera takes the more dominant element and adjusts exposure accordingly - meaning the black guy’s face becomes even darker, and now the camera probably can’t distinguish the facial features effectively. This is all compounded by the fact that lighter surfaces reflect light, and darker ones absorb it.
If you add some lighting to the setup directed at the person’s face, my understanding is that the issue goes away. When I was a pro cameraman, this was a frequent issue; we’d sometimes recommend that black interview subjects not wear white shirts as that would mean we’d have to expose for the face and the shirt would “blow out.” There are complicated lighting schemes involving scrims and such which help but that means extra gear to cart around…
Interesting on the first one, if you read his nametag it’s backwards. So either someone flipped the video, or they are actually doing it into a mirror.
I’m not sure what the point of the mirror is, because it clearly does have issues with identifying a black face as a face.
I would guess that it has threshholding for colors to recognize what is a face and what isn’t and they probably didn’t test enough black people in enough circumstances to tune it in.
There was a similar news story regarding Nikon’s CoolPix camera. In this case, whenever he took a photograph, the camera’s owner would get an error message: “Did someone blink?” No blinked, his family was Asian.
This sounds like a great idea for a sitcom episode!
Or they were recording it off the screen. Some Webcam apps (including iChat on the Mac) will mirror the video of “yourself” that they play back to you, because that’s the image most people are used to seeing of themselves. One of those little UI things that you don’t notice until something (like text) makes it obvious.
As a video security professional, I can confirm that this is a major issue in commercial video security as well.
If lights point straight down, then cameras pointing straight down will work nicely.
If you’ve got radiant light as if from a hanging lamp, for instance, coming from behind an employee manning a cash register and lighting up customers, you’re good with your cams that shoot from behind the employees towards the customer.
Few stores are set up like that.
Combine this with bright windows that either light the subject from the side or, worse, the back, you’ve got a recipe for bad headshot photos on light-skinned subjects and even worse headshots on dark-skinned subjects.
If a facility is willing to re-work lighting, I can always get you a good shot. If you’re willing to buy top-notch cams, I can mitigate your situation somewhat, but if you’ve got bad architecture and no willingness to add lights, it’s a crap-shoot.
Not the same issue, but similar. At a place where I used to work the urinal in the restroom had a motion sensor that automatically flushed when you moved away.
I found that when I was wearing my black leather jacket the thing would never flush.
That might have been a passive infrared. Some of those pay more attention to heat than to light, in which case the surface of the coat was probably closer to room temperature than your regular work shirt. As such, your movement didn’t cross the threshold.
Incidentally, I’ve got a PIR in the bathroom at my job that will sometimes trip just because the HVAC cuts on and floods the bathroom with hot or cold air.
Thankfully, said PIR operates the electric paper towel dispenser, not one of the porcelain appliances.
Does it not work with a white person in a lighter than average room, though? Glad I don’t use webcams in general–this seems really annoying.
Based on my general camera knowledge [I’m no webcam expert], it would happen, yes.
In most non-theatrical scenarios, you never get enough light to do what you’re describing.
On stage or in a studio, you can get to the lighting scenario where the white guy becomes a featureless blob and the black guy is simply very well-lit. In those situations, of course, there are professionals present whose job is to keep that from going down.
One of the professors at my university recently completed an AI research project involving facial recognition. It was larger scope (and obviously different from simple business application) and involved a machine learning HOW to recognize faces from (effectively) a blank slate and teaching it through a corpus and interaction (in an attempt to dispute the “humans are born with a predisposition to recognize faces” theory), but I’ll see if I can get him to give any input. I don’t know him personally, he just gave a talk in a research class, so I can’t promise anything, but I’ll see what I can do and attempt to get his input.