How Accurate Are Flock's AI-Powered License-Reading Cameras? (businessinsider.com) 111
Futurism reports:
404 Media revealed that a July 10 audit by the Los Angeles Police Department Office of the Inspector General caught the department's ALPR cameras generating 161 false stolen-vehicle alerts in just two months — each one ending with officers pulling over an innocent driver. Factoring in 337 alerts which "resulted in the recovery of stolen vehicles," the LAPD's cameras carry an error rate of 32.3 percent, effectively giving officers a one-in-three chance at pulling an innocent person over.
"The biggest issue remains this national database at the FBI called NCIC," Flock's CEO Garrett Langley recently told The Drive, complaining about "how archaic its structure is." Flock CEO: There's no feedback loop mechanisms; it's really just a static comma-separated file.
We've tried to build around it. We have this concept called "suppression." A local agency — let's take Atlanta, where I live — might see that there's a stolen plate out of California, but it's already been resolved because it's a rental car or a dealer car. They can, in Flock, suppress that: "Never alert us on that for the next year." That's to get around the fact that they can't force another agency to remove it from NCIC. Ideally, the way it would work is if enough agencies — let's call it one, two, or three — flag a tag as no longer valid or a bad entry, it should just get removed. Or at least it should get pushed more aggressively by the FBI... I'm hopeful that they'll see there's an opportunity to make something like NCIC, which is an important tool not just for companies like Flock, but for police departments to work together, start to make some enhancements to modernize what's a central part of our policing system.
Later in the interview he says "It's less than one in a million alerts that an officer says, 'I'm not sure if that's right.' Less than one in a million."
A recent report from Business Insider shares a worst-case scenario. In the summer of 2024 a Sacramento police officer said Flock's cameras had sent six false alerts for the same vehicle, wrongly saying it had been stolen or used in a felony crime in the nearby suburb of Roseville: Roseville police could see that Flock's software kept confusing the "9" on the man's license plate for an "8." The car owner said he would take off his license plate cover. Soon after, it happened again. It's "easier at this point we have it memorized," a dispatch supervisor in Roseville wrote in an email to a colleague.
Flock says that in optimal conditions, its cameras accurately read more than 96% of license plate characters. Hundreds of pages of records from the Roseville Police Department show a different picture. In 2023 and 2024, Flock sent 1,427 alerts to Roseville police, flagging vehicles as stolen or used in a felony after they passed one of the city's Flock cameras. An analysis by the police department found that in 71% of those alerts, Flock's machine-learning software incorrectly read the license plates... The cameras regularly missed vehicles, captured blurry images, misread license plate characters and states, and sent delayed alerts to police about vehicles possibly connected to crimes, the records show...
A factor that contributed to the misread problems in Roseville was the "particularly unique deployment" that the city requested, a Flock spokeswoman said. Roseville said it has its cameras configured so that they capture only the backs of vehicles, a setup intended to avoid capturing personally identifiable information like faces. Roseville's setup included older hardware and placement of cameras higher and further from vehicles than the company typically recommends, Flock added...
In Toledo, Ohio, driver Brandon Upchurch was mauled by a police dog after a Flock camera misread the "7" on his license plate as a "2." As a result of his injuries, he said, he lost his job and was evicted from his home. He settled a lawsuit against the city and police officer for $35,000. None of the incorrect Flock alerts in Roseville resulted in a traffic stop or arrest, a department spokesman said. Roseville requires police officers to verify that a license plate is stolen or have independent reasonable suspicion of a crime before making a traffic stop...
Flock said Roseville's camera performance has significantly improved, which the police department disputed... Flock's cameras also missed vehicles altogether... At one point, Flock's product director for machine learning suggested that officers ask drivers to remove license plate frames that made it "very difficult for the machine vision to tell it is actually a 'E' and not an 'F.'" Roseville told Flock at the time that it wouldn't do so, according to the department spokesman. Flock in 2024 shared an analysis with Roseville's police department, outlining reasons for the misreads. Vehicles far from a camera were sometimes blurry. Plates were cut off by trees or license plate frames. And Flock's software confused similar characters, mistaking an "N" for a "V", or a "1" for a "4...."
Roseville isn't the first organization to flag inaccuracies in Flock's technology. In 2021, the research firm IPVM independently tested Flock's license plate readers, concluding that Flock misidentified the state in about one of 10 reads, and that the system regularly misclassified vehicles' type and make. IPVM said that Flock subsequently blocked it from purchasing its cameras for testing.
Thanks to long-time Slashdot reader Cognitive Dissident for sharing the article.
"The biggest issue remains this national database at the FBI called NCIC," Flock's CEO Garrett Langley recently told The Drive, complaining about "how archaic its structure is." Flock CEO: There's no feedback loop mechanisms; it's really just a static comma-separated file.
We've tried to build around it. We have this concept called "suppression." A local agency — let's take Atlanta, where I live — might see that there's a stolen plate out of California, but it's already been resolved because it's a rental car or a dealer car. They can, in Flock, suppress that: "Never alert us on that for the next year." That's to get around the fact that they can't force another agency to remove it from NCIC. Ideally, the way it would work is if enough agencies — let's call it one, two, or three — flag a tag as no longer valid or a bad entry, it should just get removed. Or at least it should get pushed more aggressively by the FBI... I'm hopeful that they'll see there's an opportunity to make something like NCIC, which is an important tool not just for companies like Flock, but for police departments to work together, start to make some enhancements to modernize what's a central part of our policing system.
Later in the interview he says "It's less than one in a million alerts that an officer says, 'I'm not sure if that's right.' Less than one in a million."
A recent report from Business Insider shares a worst-case scenario. In the summer of 2024 a Sacramento police officer said Flock's cameras had sent six false alerts for the same vehicle, wrongly saying it had been stolen or used in a felony crime in the nearby suburb of Roseville: Roseville police could see that Flock's software kept confusing the "9" on the man's license plate for an "8." The car owner said he would take off his license plate cover. Soon after, it happened again. It's "easier at this point we have it memorized," a dispatch supervisor in Roseville wrote in an email to a colleague.
Flock says that in optimal conditions, its cameras accurately read more than 96% of license plate characters. Hundreds of pages of records from the Roseville Police Department show a different picture. In 2023 and 2024, Flock sent 1,427 alerts to Roseville police, flagging vehicles as stolen or used in a felony after they passed one of the city's Flock cameras. An analysis by the police department found that in 71% of those alerts, Flock's machine-learning software incorrectly read the license plates... The cameras regularly missed vehicles, captured blurry images, misread license plate characters and states, and sent delayed alerts to police about vehicles possibly connected to crimes, the records show...
A factor that contributed to the misread problems in Roseville was the "particularly unique deployment" that the city requested, a Flock spokeswoman said. Roseville said it has its cameras configured so that they capture only the backs of vehicles, a setup intended to avoid capturing personally identifiable information like faces. Roseville's setup included older hardware and placement of cameras higher and further from vehicles than the company typically recommends, Flock added...
In Toledo, Ohio, driver Brandon Upchurch was mauled by a police dog after a Flock camera misread the "7" on his license plate as a "2." As a result of his injuries, he said, he lost his job and was evicted from his home. He settled a lawsuit against the city and police officer for $35,000. None of the incorrect Flock alerts in Roseville resulted in a traffic stop or arrest, a department spokesman said. Roseville requires police officers to verify that a license plate is stolen or have independent reasonable suspicion of a crime before making a traffic stop...
Flock said Roseville's camera performance has significantly improved, which the police department disputed... Flock's cameras also missed vehicles altogether... At one point, Flock's product director for machine learning suggested that officers ask drivers to remove license plate frames that made it "very difficult for the machine vision to tell it is actually a 'E' and not an 'F.'" Roseville told Flock at the time that it wouldn't do so, according to the department spokesman. Flock in 2024 shared an analysis with Roseville's police department, outlining reasons for the misreads. Vehicles far from a camera were sometimes blurry. Plates were cut off by trees or license plate frames. And Flock's software confused similar characters, mistaking an "N" for a "V", or a "1" for a "4...."
Roseville isn't the first organization to flag inaccuracies in Flock's technology. In 2021, the research firm IPVM independently tested Flock's license plate readers, concluding that Flock misidentified the state in about one of 10 reads, and that the system regularly misclassified vehicles' type and make. IPVM said that Flock subsequently blocked it from purchasing its cameras for testing.
Thanks to long-time Slashdot reader Cognitive Dissident for sharing the article.
Cheapass hardware and poor software design (Score:1)
Image recognition can perfectly distinct between an E and F as long as the picture is good.
Now Flock crapware is working with low quality images and told to decide whether it is an E or F without the option to admit it might be either of them.
Comma injection (Score:5, Interesting)
it's really just a static comma-separated file.
Would painting a comma on a license plate (preferably one that is very visible for cameras but poorly visible with the naked eye) protect you from their surveillance?
Re: (Score:2)
Goddamn it! What's with this ticket about the CSV import being broken?!
Re: (Score:3)
Semi-colon drop tables semi-colon
Re: (Score:3)
Image recognition can perfectly distinct between an E and F as long as the picture is good. Now Flock crapware is working with low quality images and told to decide whether it is an E or F without the option to admit it might be either of them.
At a minimum, it would seem Flock should validate the tag is in fact a valid tag for state X and the vehicle matches the state's description. That would at least throw out bad reads , assume Flock wouldn't just flag it as an invalid tag so it would still generate a hit in the DB.
Re:Cheapass hardware and poor software design (Score:5, Insightful)
How would Flock know which state to which the license belongs? I would think if it has troubles reading the numbers, which are usually large compared to the rest of the stuff on the plate, it would have troubles reading the state as well.
Re: (Score:3)
How would Flock know which state to which the license belongs? I would think if it has troubles reading the numbers, which are usually large compared to the rest of the stuff on the plate, it would have troubles reading the state as well.
They have to have a way, or else there whole system is even more fucked up. States can and do duplicate tags, so a simple number search could cause problems. I suspect the whole state reading is fucked, because of the large variation in how states put the name on a tag, the wide variety of tag graphics even within states, etc.
Re: (Score:2)
Yea this. You don't punch plates in prison anymore. Now that you can mask print them, states can do anything. Since driving is also in the state's wheelhouse, all the requirement's for licensing and registration is under their laws too. This is why centralization is a pain in the US. You need a federal mandate and they have to give money for the thing or the states won't lift a finger.
Re: (Score:2)
How would Flock know which state to which the license belongs?
They have different color schemes, background art, fonts, spacing on the digits, etc. Image recognition can easily do this without having to OCR the text.
Re: (Score:2)
At a minimum, it would seem Flock should validate the tag is in fact a valid tag for state X and the vehicle matches the state's description.
The summary describes cases of looking for a "stolen plate", where a plate is transferred from one vehicle to another to obscure the identity of the owner. Even for stolen vehicles, the plates may get swapped, so you often do want to pull over a vehicle when the description doesn't match the plate.
Re: (Score:2)
At a minimum, it would seem Flock should validate the tag is in fact a valid tag for state X and the vehicle matches the state's description.
The summary describes cases of looking for a "stolen plate", where a plate is transferred from one vehicle to another to obscure the identity of the owner. Even for stolen vehicles, the plates may get swapped, so you often do want to pull over a vehicle when the description doesn't match the plate.
Yea, that's a challenge. At a minimum, Flock should do a fuzzy search on the tag with the state registry to see if a missed read digit results in an hit, but that would mean more work and cost more money. Flock's real problem seems to be GIGO compounded by poor programming logic.
Us number plates are hard to read (Score:4, Informative)
By contrast the UK and EU standard requirements are far more legible.
https://static.wixstatic.com/m... [wixstatic.com]
US pattern
https://assets.catawiki.com/im... [catawiki.com]
Re:Us number plates are hard to read (Score:5, Interesting)
That difference is just bonkers to look at. But unsurprising: the UK designed the registration plate as a standardized machine-readable piece of road infrastructure. US states treat it as a state-issued identity/emblem/mini-billboard which also happens to carry a registration number. States literally sell specialty plates for money. Sometimes a state will go for a sane design, but mostly they just arms race their way to the gaudiest shite possible, legibility be damned. And then have creepy shitty AI ALPR readers with wild false positive rates trying to pick up the pieces.
Sometimes the UK's centralisation is a really obvious advantage, and this is one of those cases.
Re: (Score:1)
Re: (Score:1)
Re: (Score:2)
Gosh, what a thing to live in a country with unstoppable child rape gangs. How *is* President Trump, by the way? Bitten any 13 year old nipples lately? Silly question, of course he has!
It is always, *always*, ALWAYS projection with you lot. Every fucking time.
Re: (Score:2)
Visiting the US and canada I always felt their plates were fairly useless even for human eyes never mind AI beyond a fairly snort distance. The characters are too small and the varying colours and designs of the plates doesn't help either.
Re: (Score:2, Funny)
Yes but you have to remember that the idea that any other country could have come up with a better standard for number plates goes against the American exceptionalism that the USA is the greatest and best and everything and it's consequential near total refusal to adopt any standard that was "not invented here" even if it is demonstratively superior in every way.
Re: Us number plates are hard to read (Score:1)
Agreed. A fine example is the metric system. The USA insists on using a system where 12in=1ft, 3ft=1yd and so on.
Where as 10mm=1 cm, 10cm=1m, 10m=Dm. Super simple. Not to mention liquid vs dry measurements; oz-lb or oz-cup.
What would make you think a country that accepts that kind of confusion would ever think of adopting a easy to read license plate?
Re: (Score:3)
That's a poor example of the point. The US system of measures is a historic legacy from it's British roots. It's more "I don't want to learn something new" than "not invented here".
Re: (Score:2)
Re: (Score:2)
The problem with that argument is that I can point to countless examples were the rest of the world have or are busy adopting superior systems in also sorts of areas that the USA has refused to adopt is busy resisting to the last on what amounts to nothing more than it's foreign so commy.
To keep it on road transport theme, I would mention road signs, roundabouts and emergency sirens as just three examples of the rest of the world using science to improve our lived experience and the USA resisting because it
Re: (Score:2)
Capitalism has destroyed license plate durability in the name of profit.
Re: (Score:1)
Re: (Score:2)
That "last 10 years or so" drastically underrates the length of the trend. It dates back to at least the 1930's. (Social Security Numbers.) It probably dates back further if I knew enough history. My guess is that it dates back to 1776.
There were and are other trends with countervailing pressure, and those tend to be stronger when population is sparse and not all land is owned. Also when processing and collecting information is more expensive.
Re: (Score:2)
It's interesting that the response to less than perfect license plate recognition is to brag about how a total surveillance state is so much easier to implement in Europe.
=Smidge=
Re: (Score:2)
Hmmm....
Given that there is little prospect of the deep state giving up its toys, the debate comes down to a question of whether it's better to reduce false positives. On THAT basis the UK/EU style wins. OTOH false positives are what allows campaigners about the system to build up a case against it. So is it better that a few should suffer in order that there is a better chance for freedom for the masses?
Re: (Score:2)
Given hat innocent people are necessarily going to get hurt, and a successful surveillance state would be much MUCH worse about that, and that innocent people mistakenly caught up will generally receive better outcomes than innocent people intentionally targeted (e.g. getting charges dismissed etc), then it seems like leveraging the flaws of the system to argue against its creation is definitely the better option.
=Smidge=
U.S. license plate design is quite bad. (Score:5, Informative)
Re: (Score:2)
Same in the UK. Designed for legibility above all else, and deliberately so
Re: (Score:2)
The UK have also removed certain characters from [A-Z] [0-9] from our numberplates as they can be too easily confused. So for example you cannot confuse a 1 and I if your numberplates simply don't have I's in them. Same for 8 and B. O and 0 etc. It does reduce the number of possible plates you can produce but with six digits in the plates there are hundreds of millions of possibilities.
IMHO I would also be an idea to include a check character, that would virtually eliminate mistakes in a stroke going forwar
Re: (Score:2)
The UK have also removed certain characters from [A-Z] [0-9] from our numberplates as they can be too easily confused. So for example you cannot confuse a 1 and I if your numberplates simply don't have I's in them. Same for 8 and B. O and 0 etc. It does reduce the number of possible plates you can produce but with six digits in the plates there are hundreds of millions of possibilities.
IMHO I would also be an idea to include a check character, that would virtually eliminate mistakes in a stroke going forward.
US plates are the same, which is why vanity plates are often a mix of numbers and letters.
Re: (Score:2)
I remember some chao driving around Whitechapel in the 90s in a souped up Merc with numberplates J11 HAD. Wonder what happened to him...?
Re: (Score:2)
The UK has the same, but mistakes do still happen - but they're usually found and dealt with pretty quick, not least because the make, model and colour of the vehicle are listed alongside the registration. It's possible there are two silver audi q4s with the number plates AB72ABC and AB12ABC (or AB22ABC, etc), but it's pretty unlikely (I don't know for sure, but I wonder if the DVLA who hand out the plates deliberately avoid such things?).
We also have a lot of ANPR (plate recognition) - you may consider it
Re: (Score:2)
Seems we prefer to cater to personalized plates. Would you like a Heart, or Hand, or Star on your plate?
Won't be long before we go full emoji.
Possible solutions (Score:5, Interesting)
1. Fine FLOCK for misreading the plates, that will very quickly make the report only the plates they are 100% sure, or perhaps even add a human in the loop for each notification - given the relatively small number of alerts given in the article, that should not be a huge number of humans on call 24/7 to confirm the reading of the plate, or deem it inconclusive (or face a $500k fine if they are wrong).
2. Any agency that doesn't invalidate their submitted license plates for alerts in NVIC and someone gets pulled over for nothing, gets to pay full restitution to the injured party AND gets all of their submissions blocked for a year on first offense, 2 years on next offense, 4 years for next offense, and so on. Competent agencies get to use the system responsibly, incompetent ones get banned eventually forever.
Re: (Score:1)
The best solution is to fire all blind police officers and hire police officers who can read and verify license plates before pulling anyone over.
Re: (Score:3)
From TFS:
In Toledo, Ohio, driver Brandon Upchurch was mauled by a police dog after a Flock camera misread the "7" on his license plate as a "2."
I hate the Flock situation as much as the next guy, but this guy wasn't mauled by a police dog because of a Flock mistake. Look elsewhere for that blame and fix that shit before we move forward.
Any agency that doesn't invalidate their submitted license plates for alerts in NVIC and someone gets pulled over for nothing, gets to pay full restitution to the injured party AND gets all of their submissions blocked for a year on first offense, ...
Agencies that have, and continue to, actively prove that innocent people get caught up and abused by their system should not get access to this in the first place.
How many ALPR's are also giving out tickets automatically? What's the acceptable false positive rate for those? (It should be zero false positives
Welcome to SlashFlock (Score:2)
News for the observed, Stuff that's captured
Re: (Score:2)
Discovery Channel has Shark Week. /. has Flock week
Re: (Score:3)
Would you like some more bitcoin or AI stories?
Re: (Score:2)
Contradictory statements in the article (Score:2)
Who and what are we to believe here? LAPD audit says the 2-month error rate was 32.3%, yet Flock CEO says misreads are less than 1 in a million.
While both statistics can be correct, they are probably measuring very different things. If any given camera is reading thousands of plates a day, then an error rate of 1:10^6 is plausible. But that's not the correct metric by which to understand the consequences of false positives in the system, because most reads--correct or not--are not associated with crimina
Re: (Score:2)
My read was less that on in a million plates are read incorrectly. Perfectly believable, the Flock cameras are probably reading hundreds of millions of plates a day. The ~â..." error rate is that the number of plates flagged that are wrong, which is partly because a small number of plates are repeatedly being flagged incorrectly.
Two things can simultaneously be correct.
Re: (Score:2)
Who and what are we to believe here? LAPD audit says the 2-month error rate was 32.3%, yet Flock CEO says misreads are less than 1 in a million.
How does the CEO know the error rate? I highly doubt anyone is hand verifying all the millions of captures. He is either extrapolating from a cherry picked subset, or quoting a failure rate of the camera to capture a plate, correct or not. E.g. how often does the camera detect a car but fail to capture a license plate.
Re: (Score:2)
Who and what are we to believe here? LAPD audit says the 2-month error rate was 32.3%, yet Flock CEO says misreads are less than 1 in a million.
IMO, the scope appears to be different. I think it's:
* Flock 1 in a million - one false positive out of all reads, including false positives, positives, negatives, and false negatives. Furthermore, that's by character, rather than by plate.
* LAPD 32.3% error rate - of the Flock positive alerts, 32% are wrong in some way (which could include errors in the NCIC database).
Flock picked the stat that appeared to be most favorable to them. Like this comment points out (https://yro.slashdot.org/comments.pl?sid=240
Irrelevant (Score:3, Insightful)
Accuracy does not matter. Legality does not matter. The cameras and the data they gather are unconstitutional - end of story.
SCOTUS has abandoned us to the Leviathan, so do your jury duty.
Re: (Score:2)
The whole point of plates IS to identify a vehicle and where and when it was. Whether thats done by eye or by machine doesn't make a huge amount of difference so long as in the latter case the data isn't stored if the plate isn't flagged in some way.
Re: (Score:2)
Whether thats done by eye or by machine doesn't make a huge amount of difference so long as in the latter case the data isn't stored if the plate isn't flagged in some way
Of course the entire point is that it is being stored. All of it., "flagged" or not. Imagine being an apologist for this shit.
https://www.yahoo.com/news/politics/articles/thanks-flock-cameras-drove-across-145634632.html
Re: (Score:3)
The whole point of plates is to identify a vehicle. But not "where and when it was". At a moment in time when the information is asked for, yes. Having the entire history of when and where it was... that's mass surveillance and should not be allowed. A "flag" in the computer can easily and rapidly be changed more or less on a whim.
It does make a difference when it is stored by a machine and becomes easily searchable. It is a qualitatively different use than a policeman noting the same vehicle during a stake
Re: Irrelevant (Score:2)
Better not drive near any normal cctv cameras or ring doorbells then without covering your plates first.
Completely unsurprising (Score:2)
The issue with trying to measure amount of false positives of classifiers giving cops pretext to act is that those false positives are largely not a bug but a feature to them.
From the statement issued LAPD OIG;
"officers acknowledged 161 alerts as accurate license plate matches; however, subsequent investigations determined that the vehicles were not stolen"
and later in the article;
The report also states that, on the vast majority of hot list hits, no action is taken by police meaning that specific people are being subjected to tracking and surveillance for no readily discernible reason.
I'd bet $10 it's something like that the beeper allows you to set it to something like "alert even on partial match" and essentially just trips on a whole load of cars all the time. And Sergeant White Horse gets to initiate stops with something better than "the vibe was bad"
good ol CSV (Score:2)
And if the comma key is broken on your clicky mechanical keyboard, there's always the Tab key. We survived the 20th century with PCs that had 64M ram. And wired connections.
So it works (Score:2)
Re: (Score:2)
CAN they be fixed? Also, is it worth the effort of fixing them to the folks that would need to do the fixing?
ISTM that the false positive rate in a system like this would be inherently very high, and no system is perfect. From other stories is appears that the system has been designed to avoid people being able to correct mistakes that they notice. This indicates that either the high mistake rate is a feature, or at least that they don't care.
They have done studies (Score:5, Funny)
just require actual officers in the loop? (Score:3)
I don't understand why they aren't legally required to keep police officers as a link in the process?
For example, if a flock camera says a stolen vehicle was spotted, it should forward that image to an actual officer that LOOKS at the image, that's given some options that INCLUDE "image quality too low for reliable identification".
And then the officer is REQUIRED to verify the validity of the warrant, and sign-off on both the identification and the warrant before action is taken. Make an officer take responsibility before sending cars after them.
Then if the plate was read wrong, or the warrant had been withdrawn or was expired somehow, they can't just waive their hand and say "too bad you got arrested at gunpoint, the camera told us to do it, it's not our fault, you'll have to try to sue the camera."
Forcing a person to make the final determination and take responsibility for the response just seems like common sense?
But I suppose it's just too easy and convenient to sit around and wait for a camera to "hit" on a car and then charge blindly at it, if you know you can just waive responsibility if something goes wrong.
Re: (Score:2)
Re: (Score:2)
There are legal requirements and then there are actual use cases. In actual use cases people will tend to take short-cuts to make their jobs easier...like just accepting the report without either checking or thinking about it.
Re: (Score:3)
I don't understand why they aren't legally required to keep police officers as a link in the process?
There IS a police officer in the loop already. Who do you think pulls over the cars? It should be a simple matter to include the actual plate number of the vehicle reported stolen in the text of the BOLO message. It's training. Cops are conditioned not to challenge authority. Which in the past might have been a crooked chief of police or mayor. But now includes a wonky camera.
Re: (Score:2)
Cops are basically a tool of the state, a hammer is the best description. It amounts to this right now:
Machine says car stolen. Arrest criminal. They don't want cops thinking too much.
How accurate is Police Officers understanding of t (Score:2)
bad policy == bad metrics (Score:2)
Oh FFS it is not AI (Score:3)
The marketing flacks tell them to call it "AI" because that gets people thinking it's SOOPER HIGH TECHY TECH stuff. Basically it's because 1) We decided that acronyms and words don't have to mean anything anymore so companies just lie through their teeth and because, 2) in the words of Henry Rollins "SOME PEOPLE WILL BELIEVE ANYTHING YOU TELL THEM!" Actually a disturbingly high number of the people will believe anything they are told. They're the same people who think that using rockets to replace airplanes is a viable idea.
Granted, it isn't like perpetual doom trollers OpenAI and Anthropic have developed AI.
Re: (Score:2)
An interference model for detecting numberplates could easily be run on a low price TPU. However even more traditional image processing using hidden Markov models would still be AI. Remember boys and girls not all AI is neural networks.
Re: (Score:2)
You're not going to claim that we've been using AI for years to do OCR and sort pictures at home, are you? In fact given that Flock is wrong (supposedly ~70% of the time) and I can verify from my own use that Ubiquiti doesn't do the best job with license plates or faces.
Re: (Score:2)
"You're not going to claim that we've been using AI for years to do OCR and sort pictures at home, are you?"
The official guide to machine vision on Arduino dates to 2020: https://blog.arduino.cc/2020/0... [arduino.cc]
Re: (Score:2)
Are you going to call the 1960's / 1970's Lizabots (the precursor for today's frontier models) AI? For me this one it would be entertaining as I've seen mimics of the systems in operation.
Do you define all fuzzy logic as AI? Oh man, this would be HILARIOUS..
Were they doing AI back in the early 1990's? I worked at a company that was building computer vision systems that could tell the difference between a human hand and a fly.
Obviously you disagree with researchers like Yan
Re: (Score:2)
"Were they doing AI back in the early 1990's?"
AI was at the level of "a course I could and did take in college" by the early 1990's. Of course at the time the fashion was expert systems, not neural networks.
"Obviously you disagree with researchers like Yann LeCun who say that LLM's are ultimately a dead end."
Since I said nothing for or against the viability of LLM as a path to future development, I don't know how any position I might have is "obvious". My personal opinion would not be very important as I am
Re: (Score:2)
But the gripping hand is that your topic thus far has been about AI, not LLMs specifically. In response to your claim that AI vision inference systems have not been in use I gave an example that does not just exist but practical to implement at even the hobbyist level.
I don't see you demonstrating it's actually Artificial INTELLIGENCE. Why? Because nothing about this is AI. Nothing about the cameras or the backend has intelligence.
Re: (Score:2)
"I don't see you demonstrating it's actually Artificial INTELLIGENCE. "
The Arduino guide uses TinyML, which can use any of a variety of neural networks, decision trees, etc.
Re: (Score:2)
"or some other bunch of science words you can string together?"
You keep avoiding saying it is intelligent, much less an intelligence. You also dodge how systems, that are not AI, were used for years to do things that just get AI slapped on them now. The developers of those systems would have laughed in your face if you called the product AI.
Re: (Score:2)
Machine learning is *by definition* AI.
"You also dodge how systems, that are not AI, were used for years to do things that just get AI slapped on them now. "
Yann LeCun has a machine vision example in his Turning Award lectures; the example is from 2011. Why do you hate Yann LeCun so much?
Re: (Score:2)
They could be running something local like OpenCV. I don't know if anyone has done an anonymous teardown of a Flock camera yet.
tl.dr (Score:2)
4% per character failure is unacceptably bad (Score:1)
If all characters had the same failure rates, that would be a plate-failure rate of about 25% per plate. Yes, I know it's not the same for each character, so the per-plate accuracy may be higher.
Until computer-only recognition is about as good as either human-only or human+computer-only recognition, you need a human in the loop.
Re: (Score:2)
I'm not really convinced that people would do any better in the same job. It's just a very poorly designed job. It's a job that shouldn't be done, no matter how attractively you can describe it, because the inherent failure rate is too high, and comes with unreasonably high potential consequences.
Re: (Score:2)
And people are lazy. At least, a computer will keep running the same poor code, over and over.
Anecdote: I received a red-light camera ticket a while back. Date and time, location and a photo from the camera. Evidently, a meat-sack mistook an 'O' for a 'Q' and got my truck, a nearly 50 year old FJ40. But the camera picture was of a late model Chevrolet. The operator evidently never bothered to look at the DMV record called up, compared it to the photo and said "Nope."
It's fine if it has a near-0% false positive rate (Score:1)
If it has a high false-error rate the city may miss out on revenue and someone with a warrant may not be arrested, but any false-positive is an injustice to the person pulled over.
I'm willing to be the one-in-10,000 false-positive from a license-plate misread but not the one-in-100 (1% false positive) or 1-in-4 (25% false positive).
On the other hand I can tolerate my city missing out on ticket revenue or a missed warrant-arrest because only half of its license plates can be confirmed as "almost certainly co
Make your mid up (Score:2)
Re: (Score:2)
1 in 3 the alert is wrong. 1 in 1,000,000 the officer acknowledges the possibility that it might be wrong.
Also, 58% of all statistics are just made up, including both the 1 in 1,000,000 statistic he made up and the 58% statistic I just made up.
Re: (Score:2)
The issue of accuracy is secondary to ... (Score:1)
Just top testing (Score:2)
That sounds so familiar [thehill.com]. Just stop them from purchasing and you won't have to worry about them testing and finding flaws in your cameras.
Accuracy Improvements (Score:2)
The Flock systems accuracy will improve by leaps and bounds if they get sued everytime someone gets
pulled over and has guns pointed at them due to an error on their part.
While the police in this country have qualified immunity and the ability to commit aggrevated assault on
anyone in the name of " officer safety ", Flock does not. Of course, you can also go after the Department
for failing to ensure officers don't rely solely on Flock data that is putting innocent drivers in danger of an
armed encounter.
Assum
12-18 months away (Score:2)
They are 12-18 months away from total perfection, and world domination!!!
Not accurate enough! (Score:2)
Re: (Score:2)
It could be probable cause if the vehicle (or the one with the similar plate) was reported stolen. If the cops were smart enough to read the actual warrant, not just respond to a Flock alert and spot the correct vehicle, then the driver could be arrested and charged with possession of a stolen vehicle.
It certainly doesn't prove who was actually driving the vehicle...
If someone gets pulled over, then the proof of "who was driving" is whoever is sitting in the drivers seat at the time.
Re: (Score:2)
ALPRs on cop cars are bad too (Score:2)
136113966 (Score:2)
So when are we getting Back to the Future-style reflective barcode license plates?
math (Score:2)
Flock says that in optimal conditions, its cameras accurately read more than 96% of license plate characters.
that's a per-character stat.
If your average license plate has, say, 6 characters, you get a 78% total probability that it reads them right.
If you have 8 Flock cameras in your town of let's say 100,000 people, sitting at busy roads or intersections, 10 to 20 thousand cars per day is a reasonable estimate, which means you get 62,400 to 124,800 mis-read license plates. Per day. Every day. Under optimal conditions.
That's more false readings than you have people.
I wonder if just randomly assigning traffic ticket
V and N, 9 and 8, 1 and 4 don't look the same... (Score:2)
Re: (Score:2)
Also: one million alerts correctly classified for every single failure? Let's say it takes the officer 10 seconds to register an alert and assign it to a local policeman for investigation. 10 million seconds at 8 hours per day uninterrupted is 348 working days: a little under a year and a half. That's 348 days of accepting the computer's choice no matter
Re: (Score:3)
More-so, it highlights the misinformation used here. They start with:
cameras accurately read more than 96% of license plate characters
... which equates to about 1 in 4 plates read wrong. Then state:
It's less than one in a million alerts that an officer says, 'I'm not sure if that's right.'
Well, that should be WAAAAY higher. They're doing an absolute shit job of catching the errors then! That stat isn't saying what they think it says. It's almost like they're not even trying and just assume it's right! Why, there might even be lots of examples of innocent people getting abused by the police because they didn't do their due diligence, like this also from TFS:
driver Brandon Upchurch was mauled by a police dog after a Flock camera misread the "7" on his license plate as a "2."
Why
Re: (Score:2)
>cameras accurately read more than 96% of license plate characters people in general don't appreciate what 96% means... that's 4 out of 100 wrong that's 1 out of 25 wrong for a typical licence plate with like 8 chars, that's 1 licence plate in 4 wrong. it's not one in a million.
More to the point, if say 96% of the plates are valid and not tied to a crime, a positive hit has a 50/50 chance of being a false positive. When looking at percentages like that, it's important to understand the population of true positives and false positives to understand the real odds of a false positive when yo have a positive result.
Re: (Score:2)
And that's just misreading errors! When you compound it with the mess that is the NCIC, plus other errors in the system, it's fairly miraculous that it ever actually manages a true positive.
Re: 96% (Score:2)
It's not miraculous. It's expected. It makes millions of reads per day. So it is millions of errors per day. One of them will match something. It's just statistics.
Re: (Score:2)
That's the problem with statistics and more importantly the reporting of statistics. It's not possible to know what it means.
Is it individual characters? Is it the plates? Is it the letters and numbers of the alphabet, 96% could very much mean the software works perfectly 100% of the time unless there's an O or 0 in the plate too, that would be 96% of the characters too. Are they talking one off? Maybe they mean errors per frame and through all frames the 96% character level error rates get corrected but 1
Re: (Score:2)
No, there ARE points of similarity.