A challenge to the security industry and to video surveillance installers was laid down by the Metropolitan Police yesterday after it was revealed that only one arrest made in connection with the London riots was a result of facial recognition technology.
Click here to view Figure 1.
Almost 5,000 people have now been arrested in connection with the Met’s investigation into the August 2011 riots, codenamed Operation Withern.
Of those, around 4,000 were a direct result of evidence from CCTV equipment, marking it an incredible success story for the technology. But at a conference held at New Scotland Yard yesterday, DCI Mick Neville of the Central Forensic Image Team revealed that only one of these identifications was made using automatic facial recognition technology. The remainder were all made by officers who together made up one of the largest investigation teams ever assembled by the force, numbering 633 at its height.
Of the 4,962 arrests made, 3,145 people were charged, with an 81 percent conviction rate for offences carried out between 6-9 August 2011.
Many of the identifications were made by so-called “super-recognizers” — people who are exceptionally talented at recalling faces.
The problem with facial recognition technology, according to DCI Neville, is invariably linked to the position of cameras in high-up places. Evidently, this type of position is crucial in recording the overall scene of a location and in viewing any criminal activity. But officers suggested that adding a camera that was positioned at eye level — perhaps at the entry point of a location — would enable much clearer images of faces to be recorded, something that would improve the ability of facial recognition technology to be an effective tool.
DCI Neville believes that businesses will drive the use of facial recognition forward. Banks, for instance, could use the technology to identify people applying for bank loans, and they might be able to automatically confirm that a person is who they say they are by comparing the image with that of the DVLA database. Neville underlined the point by holding his hands up to frame his own face and repeatedly emphasised that “we need the face.”
Insurance
“Facial recognition needs face-level cameras,” Neville added. He then challenged delegates in the room — made up of installers, council CCTV control room managers, and other interested parties — to make this happen.
PC Pat Horgan, who was in charge of Operation Withern’s viewing teams, added to the challenge by calling for better CCTV images in general. The range in quality of the footage taken from private CCTV cameras “varied enormously,” and Horgan wants to see HD images make up a bigger proportion of the images used. He added that perhaps insurance companies could offer reductions in premiums to those businesses that have a good high-definition surveillance system.
Let us know what you think about facial recognition by voting in our latest poll.
Related posts:
Free Download: The Video Surveillance Report 2023
Discover the latest developments in the rapidly-evolving video surveillance sector by downloading the 2023 Video Surveillance Report. Over 500 responses to our survey, which come from integrators to consultants and heads of security, inform our analysis of the latest trends including AI, the state of the video surveillance market, uptake of the cloud, and the wider economic and geopolitical events impacting the sector!
Download for FREE to discover top industry insight around the latest innovations in video surveillance systems.
“The problem with facial recognition technology, according to DCI Neville, is invariably linked to the position of cameras in high-up places. Evidently, this type of position is crucial in recording the overall scene of a location and in viewing any criminal activity. But officers suggested that adding a camera that was positioned at eye level — perhaps at the entry point of a location — would enable much clearer images of faces to be recorded, something that would improve the ability of facial recognition technology to be an effective tool.” It is all well and good positioning cameras at a… Read more »
So, the camera needs to be up to the task as well, is that what you mean?
In essence, yes. The camera can be HD quality, but there are facial recognition programs/software available that actually do the job that is required.
I think that putting megapixel cameras in more convenient places partially sorts the issue. Perhaps it will be enough, but a crowd full of people may be too difficult for the camera alone to detect faces.
I’m surprised with caps/ hoods and camera positioning as previously mentioned, that facial recognition delivered at all.
I wonder about the business use of facial recognition. Surely there would be data protection/ privacy implications here? would banks need to declare they have a camera that will be checking your face for identification purposes when a prospective client enters the building?
That is a good question, I would have to guess that it would depend upon the way they handle the recognition like say if they keep a picture of your face on file then yeah there should be some privacy notification out there but if they are just identifying a pattern based off of your features through some type of algorithm then maybe not. It was impressive that they were able to make this arrest.
Thanks, that makes sense.
I’m intrigued by these “super-recognisers”. sound like there’s huge potential in using these skills.
I think positioning of camera is one of the most important aspect need to be taken care of. I believe that there are definitely advantages and disadvantages of whatever position we are taking as far as siting of camera is concerned. The camera must be high resolution with enough optical zoom capable might be few of the choices we need to think upon. But I believe the purpose for which it is to be placed will drive its positioning.
If you’re worried about privacy implications, in my experience you have nothing to fear with the current state of facial recognition software. Whether it’s the cameras, software, or another element not mentioned here, environmental conditions (shadows, shafts of light, etc.), there are far more stories of failed facial recognition than there are success stories. I’m not negating the value of cameras that catch images straight on — we routinely use them. In fact, our first line of defense when budgets are tight is to place a camera at every entrance to a facility, catching people on the way out and… Read more »
Well, indeed, the more pressure you put the camera under the harder it’s going to be to ensure a successful identification, but the use of it would depend on the perceived/calculated maximum traffic flow. Everything has a breaking point, I’d be interested to find out what the max that an average facial recognition software can count is
I think you’re right that such face-on cameras are good practice full stop for getting a better image of a person. Who needs an image of the top of someone’s hood when you could get a straight-on picture of their face. As you say, such images are useful for human operators as well as the computers!
Although I agree with your comment, would that particular camera then be installed knowing it may not capture people’s faces? Cameras on that scale will be put up with a specification behind it, and if they want facial recognition then a camera “positioned better” may not fulfil that brief.
There’s always data protection to be aware of. I’m not sure what the specific (if there are any) rules are around facial recognition technology, but I would imagine it’s similar to general surveillance requirements: ensure that you store the image securely, with controlled access only to approved people, and that you state how long and for what purpose the image will be stored.
I think that although most would agree that angle is useful it’s often hard to implement and when you’re being cost conscious that camera covers such a small area that it’s often the first view chopped when trying to stay within budget.
Well in that case perhaps a user should be looking again at their operational requirement and see how they can better match it to their specification. What your budget will allow will ultimately guide you, right?
Right, SafenSane, angle definitely matters. It then follows that camera installation and placement matters, because how you install it affects the picture that it captures.
You’re right, it can be very site specific. Depending on ceiling height, width of the corridor, etc., sometimes you can get very good coverage with a 2-3MP camera and a good wide angle view.
Sometimes you have to look at what has more value; documentation of the incident but no identification of the perpitrator, or identification of the perpitrator but no video documentation of the incident. Often we find the latter of more value as the former may have witnesses as well. Witnesses are horrible at describing people but not so bad at describing events…
One thing is plain enough that it is actually the positioning and imaging quality of the surveillance that is in question and not the Facial Recognition Technology. But I personally think that facial recognition algorithms may also be improved to better match the distant facial features or by zooming them in without breaking the pixels.
@ JonathanL, your guess seems probable. Moreover, if identifying the person by comparing his/her image with DVLA database is the only reason for recording the face picture, there may not be any need of declaring it explicitly. Because once they have verified the identity of the person, they don’t need the picture anymore. They will have to declare it otherwise of course.
Camera placement matters greatly but all to often you get issues that you just can’t work around. Things like shelving and large items on those shelves in a warehouse for example or in a retail setting think about how much stuff is usually jammed around the counter of a convenience store. I’m using them as an example because we see footage from them very often. Nine times out of ten the camera angle is high and pointing down at the counter. Not the best angle to get a view of someone’s face but good for making sure money is passing… Read more »
In such a case, surveillance is likely to work well only with a vigilant member of staff on duty. And no matter how vigilant one is, you can’t pay full attention 100% of the time.
Yeah, those algorithms are going to get better and better and better over time. But without a clean capture of a face, it’ll be relatively useless
Having been into places like this at 2AM, I have to agree. This is usually when a store clerk is stocking shelves or other tasks that can’t be done during the busiest hours. I’ve also seen some places where even if the clerk was watching my every move the lay out of the store makes it impossible for them to see a customer in some areas. Even the best places around have blind spots.
You have to do a cost-balance with retail loss prevention right? How much merchandise can you afford to have stolen versus investment in security?
You may be interested in this related article: DNA techniques could transform facial recognition technology, https://theconversation.com/dna-techniques-could-transform-facial-recognition-technology-86027