Reminds me a bit of those drones in "They live"...
"Mamma don't like tattle-tails... BLAM !... "
A drone surveillance system capable of highlighting “violent individuals” in a crowd in real time has been built by eggheads. The artificially intelligent technology uses a video camera on a hovering quadcopter to study the body movements of everyone in view. It then raises an alert when it identifies aggressive actions, such …
Have an upvote! Watched this again a few weeks ago.
It was a mate put me onto it a good twenty years ago - we used to deathmatch Duke Nukem 3D over a serial cable with back-to-back PC's and when Duke muttered the immortal line about coming here to chew bubblegum and kick ass, and being all out of bubblegum, my mate told me it'd come from this film
This technology doesn't belong on a drone.
By the time a drone is qualified for police / military use, it is an expensive bit of kit, and will have at least a couple of operators monitoring it in real time. They will be much better at picking out suspicious behaviour than this AI.
This tech is interesting, but would be much more useful on fixed CCTV cameras, where it could be running 24/7 across thousands of cameras. The UK's city-wide camera systems, for example, could pick out fights outside bars automatically - that might actually be useful.
I suspect this is the new "on the internet" of patents. Do something interesting (novel analysis of a video stream)? OK, nice. Put it on a drone? Wow, where did all this cash come from?
"I suspect this is the new "on the internet" of patents. Do something interesting (novel analysis of a video stream)? OK, nice. Put it on a drone? Wow, where did all this cash come from?"
As far as I can tell that's exactly what it is. The introduction refers to a variety of previous similar work using fixed cameras, then goes on to say:
"Despite their impressive performance (more than 90% accuracy), the area these systems can monitor is limited due to the restricted field of view of the cameras. The law enforcement agencies have been motivated to use aerial surveillance systems to surveil large areas."
Quite literally, "There's nothing new here, but we've put it on a drone.". And of course, what it rather fails to note is that it's trivial to cover a city with thousands of security cameras, but rather less easy to do the same with a swarm of drones hovering 2m above everyone's heads.
"fixed CCTV cameras, where it could be running 24/7 across thousands of cameras."
And that, children, is the whole point of the exercise. Once those cameras are in place it will be trivial to store, share, sell, search and categorize every image they collect. Mealy mouth assurances of deleted data and "privacy protections" are only "this won't hurt" lies used to calm the victim.
This tech is interesting, but would be much more useful on fixed CCTV cameras, where it could be running 24/7 across thousands of cameras.
How would that help when the ratboys are running riot on mopeds, and the plods don't have clue what to do about it? Better video footage to sell to the news websites?
Whether false positives are a problem depends what you do with them. If an airborne Laser zaps the unfortunate victim to a greasy smut, that's a problem. However if the false positive then goes to a further level of checking and is discarded without the subject ever being aware of it or the misidentification recorded then its no big deal.
This system would definitely have a problem with a Headbanger's Ball then. The idea of drones might work for outdoor events but indoor events would require a multitude of cameras. I'm just a tad skeptical about the whole thing and living in a world of constantly being watched. Then again, the 5-eyes and everyone else has ramped up that bit of paranoia. The question is: will the average person care that they're being watched? If Facebook, etc. are any indication, the answer is "no".
"What was this person doing that was classified as violent behaviour?"
Punching, kicking, strangling, shooting or stabbing. The article is, as usual when reporting on science, rather misleading. As mentioned earlier in the article, all images contained between 2-10 people; there were none with only a single person. The classification accuracy (shown in table 3 in the preprint), is for the number of people engaging in violent activity in a given image. If a single person in a group throws a punch, the system is apparently quite good at identifying that it was a punch. If a big brawl breaks out, it's not as good at working out exactly what each individual is doing. So it might be decent at detecting something like a random mugging on an otherwise peaceful street, but would be much worse at detecting a fight breaking out on a dance floor.
What does that look like?
How often does that happen?
How many drones at 16 metres apart would be needed to keep Londoners safe from kidnapping?
As already mentioned, dancing could easily be mistaken for fighting or even a couple snogging.
I also wonder if in places like Glasgow and London, it can distinguish between a Glasgow kiss and a genuine kiss.
when it identifies aggressive actions, such as punching, stabbing, shooting, kicking, and strangling, with an accuracy of about 85 per cent.
If it's as accurate as various police forces' attempts at facial recognition, the majority of arrests will be pensioners doing Tai Chi.
Tai chi might fool the system right up to the point where you give someone a smack in the gob, there is no way you can really disguise that.
All of the 'floaty' stuff is about concentration, contemplation, building muscle memory and strength, stamina and balance.
You could probably get away with some of the joint locks and throws though.
The system is able to differentiate between Dancing and violent activities as the pose of both individuals is used for identification of the violent individual and not just one individual.
The drone can fly higher as well and can identify more individuals (greater than 10) as mentioned in the paper. It was restricted to 10 as we were trying to run it real-time and beyond that the processing was slow. That can also be managed with more resources.
"Any idea what the accuracy is at an altitude of 50 meters over a packed football stadium ?"
I think the authorities would probably drop the drone idea when the winning Cup Final goal was scored in the last minute of extra time - and the hovering armed drone slaughtered the whole crowd .
IIRC, in "Cloak of Anarchy," Larry Niven wrote about a gadget called a "copseye." It floated above crowds and had a camera and a stunner. They patrolled "free parks" in which there was no law except "no violence." If violence broke out, the copseye would stun the participants.
Then a clever guy found out how to short them all out - but his attack also caused the exits to all lock, so things got a bit interesting for a while as folks got to try anarchy for real.
So... how much work would it be to train it to recognize other suspicious actions, people trying to evade detection, people walking guiltily, "loitering with intent", protesters, people working their way purposefully through a crowd instead of gawping? I'm sure you can come up with more. It's fun!
In this case, an overly sensitive system could produce false positives for people playing football together, and think that they were kicking one another.
This is going to be fun. Lots of marginalized Dindu Nuthings suddenly showing extreme interest in various sportsy outdoors activities. No-win no-fee lawyers are reserving billboard time.
"...aggressive actions, such as punching, stabbing, shooting, kicking, and strangling."
Sounds like an average day at the park with my 5-year old nephew. If his friends are with him? Add any other sort of mayhem.
But in an ordinary crowd, I have no confidence in any surveillance technique other than trained observers with mirrored sunglasses and walkie-talkies.
Biting the hand that feeds IT © 1998–2021