When an experiment with existing technology does a “good enough” job | Hacker News Reader

When an experiment with existing technology does a “good enough” job

medium.com

223 pointsxelfer92 comments

Imagine a passive system scanning fellow motorists for an abductors car that automatically alerts authorities and family members to their current location and direction.

That's the same argument that police departments use "Well we'll only use it to capture child abductors and stolen cars and stuff".

But the truth is that there just aren't that many real child abductors out there (especially with known license plates -- and when they are known, it tends to be a child-custody dispute rather than a stranger-abduction), and meanwhile a huge database of every driver's whereabouts is being amassed with no good controls over who owns the data or what it's used for.

I'm surprised companies aren't offering the service for free in return for the database. There must be huge marketing possibilities... John Doe drives past this oil change place twice a day, let's send him a coupon, or better... John Doe's car goes to Jane Smith's house twice a week and parks there for two hours, let's send his wife an ad for a private investigator.

Oh dear. That's so creepy someone might actually do it!

I'm getting into home automation at the moment, so a use case that I think I deleted from the article was just letting you know that your mum or someone had pulled up in the driveway. License plates vs. a known list of visitors. Or as is happening here in some of our suburbs, "Alert: A group of 4 men have arrived in a stolen V8 Commodore"

The FBI basically did this where they amassed hearsay about the important people in groups they didn't like then mailed anonymous letters to their partners, friends, family, etc. The term you should Google is "cointelpro".

IIRC recently there was a city in CA that mailed everyone who drove down a certain street a letter in an obvious colored envelope saying "Dear, Whoever, consider yourself informed that prostitutes go here, kthanksbye". This is a public road that people drive down to get from A to B.

> That's so creepy someone might actually do it!

It's only creepy if it's done on a targeted, individual basis, which it won't be.

As it is - in many countries you legally have no expectation of privacy in a public place, so I have no problem with en-masse ANPR. If the police eventually develops a very dense sensor network that can track every car's movement across every intersection then there would be too much information recorded for me to personally worry about it - and I still believe it would help with crime in general. The usual reactionary bullet-points based around the usual bogey-men of terrorists and child-abduction are weak and I think distract from more meaningful and impactful, but sedate arguments: e.g. it would reduce the need for inherently-dangerous hot pursuits to follow a vehicle as the cameras would do that anyway. Uninsured vehicles could be identified immediately before they get stopped for a broken taillight or involved in a collision. And cynically: APR systems could be used to compute a car's average-speed between points to determine if the car broke the speed limit).

I don't feel mass ANPR is comparable to other bulk-surveillance schemes like Internet snooping, because I believe that we do have an expectation of privacy regarding what goes through our home connections, and acting on Internet surveillance is inherently subjective - whereas ANPR can indicate if a car is known to be stolen or uninsured, that's not something that's open to interpretation.

it would reduce the need for inherently-dangerous hot pursuits to follow a vehicle as the cameras would do that anyway.

It wouldn't really reduce the need for the infamous hot-pursuits because often in these hot pursuits, the car is stolen so the driver is not who the car is licensed to.

Hot pursuits are not primarily about identifying the driver either - it's about following the car so they know where it is so they can apprehend the driver. With a pursuit there's a panic, a high probability of the car being damaged or destroyed, and possible injury or death for the driver, the police involved, and of course, any pedestrians caught up in the way.

With a large enough ANPR network, there is no need to chase the car - so the driver won't necessarily panic or act irrationally to evade the police - leading to a higher probability of recovering the vehicle in good condition, and a successful non-dramatic apprehension of the suspects as it makes it easier to catch them by surprise, for example.

Right, it's not about identifying the driver, it's about apprehending the driver.

I hope I never live in a society where the driver of a stolen car can expect to be apprehended no matter where he stops because he'll be tracked by a camera network so ubiquitous that he can't escape it.

Because such a ubiquitous network means that all of my travels are being tracked and recorded too - which will likely include facial recognition, so there really will be no escaping Big Brother... and I don't trust any government with so much power and information over citizens to be benign.

Conversely, such is prohibited in my State. If it is to be done, it must get judicial approval, be for a limited time, data deleted, and target a specific subject.

I helped the ACLU stuff the envelopes. I consider that some of my best work.

The problem are:

- who has access and are they using it for official business?

- what if it's hacked?

- who is it shared with and who do they share it with after that? private companies for analysis?

which it won't be

All of History says otherwise. All surveillance mechanisms have been used to individually track dissidents, personal enemies, ex-spouses, etc. What makes this different?

If the police eventually develops a very dense sensor network that can track every car's movement across every intersection then there would be too much information recorded for me to personally worry about it

The availability of large databases (ex. DNA) have been shown to increase the number of innocent people caught in the net, since the probability of false positives increases proportionally and cops start relying on running wide searches over regular investigation techniques.

it would reduce the need for inherently-dangerous hot pursuits to follow a vehicle as the cameras would do that anyway

Only if you have a complete panopticon, with no gaps, which in a wide enough are will be difficult, if not impossible.

I worked for a company that did exactly this. It provides security for the majority of the gas stations in the country. It scans every single license plate of cars entering the premise. It then queries a freely accessible, government provided database for the car's model and whether it was reported stolen. If it was reported stolen, the alarm goes off. Also, if the license plate is on the country-wide blacklist that the company maintains, weeeh, the alarm goes off and the pumps get blocked. They also attempt to match license plates to individuals so they can be recognised when they return. Of course, all of this is linked to transaction history etc.

Any project in a heavily regulated environment is not measurable that way.

The cost comes from compliance, audits, paperwork. Trials, PoCs, etc.

You also need to prove security and a long term lifecycle for the solution. Those open source packages - are they being updated? Who validates patches? etc etc.

Those are moats, once you're able to deliver the above, you're set for a long time.

Good luck!

You forgot Celebration Dinners!

So just add 20 bean-counters to your team and you can bid for the other state contracts :-)

Not quite. You'll also need a minority owned business to act a pass through entity.

You're very right.

At a quick glance, this thing is going to need power, probably from the car battery. You don't want it to catch fire and kill the occupants, so there's going to be R&D in power supplies, material etc, along with regulation and testing.

Sure, if there never was something before hooked to a car battery. Or if the material would have to be kryptonite. $86 Million is just a way way way to high number for such a project. Of course the 57 lines here won't do it in production and I acknowledge a certain need for R&D but this number is just too high.

Reading plates is likely a small fraction of overall costs. Consider, with 220 people cars in a trial your going to need someone to write a manual / answer help desk questions / etc. which have nothing to do with the actual device but do add costs.

As to the device it needs to do more than just read licence plates. It must compare them with something which means it needs to get over the air updates for a list of licence plates. So, now you need some back end system to get important lists of licence plates from various other systems. As to the device, how much coverage does it get aka does it need multiple cameras.

And on it goes.

Emergency services vehicles run so much equipment they have isolated power from engine power.

I have noticed here in California the the police always leave their patrol cars running when parked. I was entering a restaurant a saw a patrol car running. I saw the officer inside and asked him why they leave the cars running. I asked him if was to be ready to take off fast.

He laughed and said that there is so much electronics inside that the battery would run down during lunch.

I didn't think to ask why they didn't turn off the electronics. I also didn't ask what they do at the end of the day.

Boot up times for the computer(s). That takes a while, though I imagine it is getting faster.

Huh there you go! thanks

Or you know, make the software for smartphone?

Oh come on. Not all heavy-handed regulation would pass a cost-benefit analysis. Nor does it cost 80 million to run some POCs and QA on a project like this.

At nearly 400,000 AUD per vehicle, the question should be what better opportunities to help people did the Australian Govt pass up in order to engage this project?

I think the point was that the costs are justified by the burden of meeting regulatory requirements, not the benefit of the regulation.

That is understood. My point was that this kind of spending on a project like this should beg the question, "what better opportunities did govt pass up to engage this project," rather than responding w/ complacency; i.e. 'that's just the cost once you factor in compliance.'

Once you start taking into account audit and recordkeeping requirements, you will quickly run into that figure.

As an aside, the trials and PoCs have already taken place as a separate project worth only $100,000.

The government is fairly corrupt. This is a make my friend rich project. See the Myki card for another example.

I don't know about Australia, but this is completely true in Central Europe. An average Joe cannot tell how much an IT project can cost, so IT projects are the best place to "make your friend rich". See the slovensko.sk project (https://spectator.sme.sk/c/20429221/governmental-agency-plan...).

> Although, the solution would occasionally have issues with particular letters.

A mistake like this can and has destroyed lives.

http://thisiscriminal.com/episode-18-695bgk/

> Police officer John Edwards was patrolling a quiet neighborhood in Bellaire, Texas when he saw an SUV driven by two young African-American men. It was just before 2am on December 31, 2008. Edwards followed the SUV and ran the license plate number. His computer indicated that the SUV was stolen, and Edwards drew his gun and told the two men to get down on the ground. It wasn’t until later that he realized he’d typed the wrong license plate number into his computer. He was off by one digit. By the time he realized his mistake, one of the men had already been shot in the chest at close range.

I'm not sure the license plate scanner is the one to blame here. Since when does stealing a car carry a death sentence?

Exactly, the fault lies with the trigger-happy police, not with the software.

Of course. But it is precisely this "trigger-happy" nature which makes an introducing a system with an appreciable error rate dangerous. This at least partially (partially) accounts for higher costs.

I wonder what the comparison of error rates between simple ANPR and "cops typing plates in" looks like...

isn't it because the officer type wrong license? i thought the scanner purpose is to minimize human error like this

An automated scanner may be more accurate, but given that it will perform far more scans, I could imagine the total never of false positives being higher.

s/never/number/

sure, he typed wrong number and got same make, model and color stolen vehicle, also:

http://www.independent.co.uk/news/world/americas/policeman-b...

This sort of thing doesn't really happen in Aus. I could see someone getting arrested and spending a night in jail unnecessarily or maaaaaaybe tased (and to be clear, whether we consider this a fair price to pay is a different question), but police rarely shoot people here.

Yes it does.

Just the other day the trigger happy cops shot a couple of people at a swingers party.

"The other day" = almost two months ago. Also, apparently (this is one of those stories with three sides) he was holding a fake gun without an orange tip or similar. Police say he was brandishing it, others say he didn't even have it on him. Who knows.

This is what I get for not following the news. I also remember a case years ago on Bondi beach where a guy having a psychotic break and brandishing a knife was killed. If I remember correctly this was used a justification for introducing tasers.

I should have said rarely.

A knife in the hands of a psycho is much more dangerous than you might think. If he's at close enough range a shooting reaction is correct.

You don't need to shoot to kill. You only need to disable a attacker, not kill him. "German Police shoot a knife wielding man in the leg in self defense"

https://www.liveleak.com/view?i=cfe_1440199540&comments=1

Anybody that shot a handgun at the range will tell you that hitting a small moving target like a leg is not a guaranteed shot even by an experienced shooter (add stress, low visibility etc and this becomes even trickier).

If you come at cops with a knife I think it's "reasonable" they shoot you. What is not reasonable in the case of US cops is that they seem to shoot even non dangerous people and get away with it.

And yet UK police don't tend to shoot people with knives.

This is also very true, and it makes a difference.

Shooting is using deadly force. Shooting a suspect to intentionally avoid fatal wounds is a TV myth (most episodes of Quinn Martin's "The FBI" ended in this way, for example).

Every LEO in the country is taught to shoot only when deadly force is necessary and to shoot for the center of mass (and continue shooting until the suspect is down).

Shooting with intent only to inflict an immobilizing but not-life-threatening wound is inherently wrong and has been held actionable by the courts. If the LEO had the luxury and time to even consider such a tactic, then aim and fire, then by definition no innocent life was in imminent danger, and therefore the use of deadly force was excessive.

And if you take another look at your source, you'll see that there is no stated intent to only wound the suspect in that German case.

A good study of such situations is the Massad Ayoob book In the Gravest Extreme.

In my American government class in high school in Texas in 1984, I was taught to behave in such a way - keep hands on wheel, do exactly what you are told to, don't move suddenly, to minimize the chance police will kill you, because they don't know if you are a criminal with a weapon or not and a police officer has to be prepared for either situation. It helps immensely to not be an ethnic minority - representations in popular culture in America alude to the life or death possibility of every police stop for those communities like the recent episodes of the TV shows Insecure or Dear White People or Bruce Springsteen's song, 41 Shots. The other side of the coin is America extremely rarely prosecutes a police officer for killing someone, no matter what the circumstances, even if there is video.

"Tuttle? His name's Buttle. There must be some mistake."

"Mistake? Ha ha, we don't make mistakes."

https://www.youtube.com/watch?v=6OalIW1yL-k

But the bureaucrat(?) did say "Buttle".

Well I don't want to spoil the movie. Brazil is worth a watch.

It's more complex than that.

Despite the tragic result, the officers were unanimously acquitted of any criminal wrongdoing.

https://www.usatoday.com/story/news/nation/2014/05/24/suprem...

"...and Edwards drew his gun" - I think this is the situation where the problem arose, not the license plate scanning itself

No 100% accurate license-plate reader exists - and that includes human eyesight. Showing license identification confidence levels, as the author's code does, is far better at dissuading the cops from over-reliance on the tool, than pretending no mistakes will ever be made.

Author here, hi HN/Reddit.

I had hoped I'd made it clear enough, but by no means am I recommending something I cobbled together in an hour on the couch be rolled out as a competitor to BlueNet. I mean shit, I'm web scraping their database and using compressed footage from cheap dashcams. The algorithm is also untrained.

I was, on the other hand, hoping to promote healthy discussion and pose some good questions related to IT procurement here in Australia. There needs to be a happy medium between what I've wired up and an $86M solution.

For a full disclaimer, many many years ago I did work on various Victoria Police IT projects in a previous company. While our projects were delivered on time and on budget, we did hear some horror stories about what the multinational consulting firms were charging.

Great article, loved it.

I worked in government once and could not believe the cost required to get what I saw as small projects done.

In the end I understood, when I came to understand, as other in this thread have pointed out, that there's support, rollout, training, documentation, project management, planning and it goes on and on. The technical solution can be just a small company of providing a sustainable solution to an organization.

Having said that, you should be packing this up and selling it.

I've worked with state government and fortune 50(0)s, so I get where you're coming from. Costs that are just part of doing business seem like enormous waste to outsiders. That being said, I've heard some horrible war stories. NDAs but.

I've got a war story about a government project we did for which we invoiced in total ~$80k, which by the time it made it to the media was a "$3.5 million dollar project".

Feasibility studies, hiring experts to write reports, the salaries of everyone in the department, etc? I have no idea, but conceivably if one wanted a juicy headline, there's ways to get it.

Also lying. If there's no comeback, then there's no real downside to heavy exageration.

> I was, on the other hand, hoping to promote healthy discussion and pose some good questions related to IT procurement here in Australia. There needs to be a happy medium between what I've wired up and an $86M solution.

A related story for this discussion: QUT students design a $500 cloud-based census server four times better than IBM’s $9 million system [1]

[1] http://www.news.com.au/technology/online/qut-students-design...

"four times better"

I have two objections, which the article doesn't seem to fully address.

First, do we know that the server the students designed actually met the spec handed to IBM? Often times a lot of the complexity of a project comes from the interaction of a few features. It's quite possible that the version the students provided didn't actually do the hard stuff. We know from the article that they specifically didn't address any security/privacy ramifications of sitting running in the cloud.

We also don't know how many similar projects were attempted and failed. Sometimes things just come together, way quicker and cheaper than should have been expected.

I'm not endorsing the work done by the student, I just thought it was relevant to the discussion. I agree completely that they probably haven't addressed all of the requirements or gone through the rigorous testing and verification required. Not to mention that when things go wrong you want somebody you can hold accountable.

To me the problem is with out a true project plan and an extensive list of requirements it's impossible to tell anything.

Let's assume that your system is wrong 1 out 10 times or 100 times. And they're are 1 million scan a year. that's going to result in 100,000 people pulled over wrong. For a cost of let's say 860K(assume your outfitting the system, training people, data storage etc.).

Let's say your going to need 4 to 5 sigma better performance to get that. Does that mean 10x more effort? 100x more effort? Think data collection, storage, possibly state of the art algorithm...

what if the data gets hacked.

All I got out of your article is there's already no way to tell the difference in cost between an toy system and enterprise, government system.

BTW, it was still a cool toy project...

I feel like there's a killing to be made in "bidding for government contracts as a service". Plus you'd be doing A Good Thing (at least in theory)

Obviously complicated requirements exist but I think there's a way of embracing this with some skilled workers

I think that the digital service is trying this a bit

There are already companies that do something like this. Carahsoft is one that I dealt with in a previous career. You sell through them to the government.

In much (all?) of Australia all government tenders have to be publicly advertised so this is something you could definitely look in to if you were interested. In particular, Victoria's listing is at : https://www.tenders.vic.gov.au/tenders/index.do

I read an interesting article earlier today about IT spending in Australia: http://www.abc.net.au/news/2017-08-28/federal-governments-$1...

> It found public servants were too afraid to make major changes to IT procurement and were not talking with other departments to avoid duplication.

> "A fear of external scrutiny of decisions — such as through Senate estimates and audits — leads to a low-risk appetite and a culture where it is 'not OK to fail'," the report said.

> "This means that old and familiar ICT solutions are preferred to newer and more innovative, but perceivably riskier, solutions."

Nobody in the APS gets fired for hiring IBM, and nobody in IBM gets fired for screwing up major government projects. http://www.abc.net.au/news/2016-10-25/turning-router-off-and...

While I agree, Aus has also had some success with IT projects.

Many federal projects are now under a single sign on portal that works very well. Our taxes take less than an hour to fill out as most details are pre-filled.

And speaking for NSW, the RTA/RMS/Whatever-they're-called-these-days (our equivalent to the DMV for US readers) is almost fully functional without an in-person visit.

Don’t ask me why, but one afternoon I had the desire to prototype a vehicle-mounted license plate scanner that would automatically notify you if a vehicle had been stolen or was unregistered.

Don't ask me why, but something tells me there was a bit more to the actual Victoria Police project than just the image-scanning component. But the title makes for good clickbait, at least.

Particularly relevant now that the Australian Federal Government has come out and said that "the cost of Government IT has jumped to nearly $10 billion.": http://www.abc.net.au/news/2017-08-28/federal-governments-$1.... And you wonder why?

$86M does sound high, but glad to see the author at least acknowledge that some of this would likely have been spent on improving existing software/databases etc to support the new system. Also probably worth noting is the cost of the labour required to develop the new system, and upgrade the old, and design and fit out the interior of the vehicles, and to make sure it's all done in compliance with privacy laws, and so on.

Not that I'm saying $86M is justified - rather that I can start to see how things might add up.

> I would expect part of that budget includes the replacement of several legacy databases and software applications to support the high frequency, low latency querying of license plates several times per second, per vehicle.

Well, if we're just checking to see if a car is registered stolen, that's hopefully a very small database. A simple hash table should be sufficient for a yes/no decision. But God only knows what goes on inside the DMV, and getting that data out into a rapidly queryable state could be tricky.

I would have thought the number of known-stolen plates would be small enough you could push the entire list out to the device.

Google reckons there's only 4.5 million registered vehicles in Victoria (.au) - with 6 character [A-Z0-9] = 36bit/5byte plates that's only gonna need ~22 meg to store them all and you've got a few bits for "flags" - even if 10% of them are listed as stolen at any time, you'd only need a couple of meg for the "stolen car list" - even without handwaving away the overhead for the data structure, this seems a reasonable solution. If you want the make/model/colour info you could then just query the remote db when you get a hit for a plate with the "stolen" flag set (or perhaps the "vehicle of interest" flag).

Of course nobody's gonna build it that way, because then you lose the data set of plate locations you can own without needing to ask for consent. This will, of course, store GPS location and timestamp of every plate is acquires, and if they're being particularly audacious, the actual imagery captured as well. That's how you make this a $80mil project... :-/

(Of course private industry already has this - tow trucks carry alpr rigs and sell the data to repo agencies in the US, I'm told...)

The local populace has a $170 million/year problem of stolen cars. Government spends $86 million (one time cost? estimated cost over X years? it's unclear) to help solve this problem. Doesn't seem like that bad of an economic tradeoff on the face of it, really.

I can understand that the $86 million seems like a lot until you realize the kinds of capabilities that an automotive camera system like has to comply with and all of the integration needed with other software/systems/services and the fact that this cost is likely a total cost over many years as it sounds much more sensational. Just the initial report from Deloitte cost the government $115,000.

In their first trial in 2014-2015, which lasted 15 months, with only 6 cars outfitted, ended up identifying and impounding 240 stolen cars and increased revenue from additional tickets by hundreds of thousands of dollars per month. That's pretty impressive! Now they're rolling out to not just 6 cars but to 220 total, which in theory should improve policing efforts.

References: http://www.caradvice.com.au/350313/victoria-police-to-consid... http://www.heraldsun.com.au/news/law-order/victoria-polices-...

Yup. I did something similar with a RasPi camera and OpenCV. On my house it store and record every license plate that drives by and/or stops. It certainly leads to some ethical questions (not that you couldn't record this data going all "Rear Window" on the neighborhood but it's quite easy to understand all the coming and goings of your neighbors.

I wonder about the cost of licensing patents for a project like this one.

Amazing and super sad that a government contract could ever be submitted for such an embarrassingly high amount. I hope more people buying gov software contracts read this.

When reaching 90% completion, like an MVP, those remaining 10% will take much longer. I also get angry when tax money is seemingly being wasted. What should you do though ? Offer to do it for free ? I'm a developer and I like building stuff, I do not like selling stuff. Thankfully there are people that actually like selling, so I suggest finding such a partner and be prepared to spend five years completing the last 10% and hopefully make some money.

This is fantastic!

I honestly believe it is examples like this that clearly show how quickly the world is changing, and how massive organizations are a thing of the past.

You don't need thousands of employees and $86M, you need a couple of smart people and a weekend.

Uber and Air BnB have to some degree done this already, I have no doubt the exact same thing will happen in health, education, transportation and many, many other aspects of our lives.

You can't say it's expensive without knowing the requirements - this is not only misleading, but it says something negative about the overall level of discourse we have.

"Why did this system cost 86 million dollars" would be much more informative. It could be that the whole thing is a waste, but that would require an actual analysis.

You could now integrate auto gun aiming on the felon and give them 20 seconds to comply.

How did you get to the number 220 of vehicles?

It was sourced from an article. I originally built the tool to see if it could be done (given I knew the DB existed), and then did some Googling and found BlueNet. I wasn't actually trying to do it better/cheaper, it just so happened it had been done before.

> Imagine a passive system scanning fellow motorists for an abductors car that automatically alerts authorities and family members to their current location and direction.

Seriously? You jumped the shark from interesting tech, to think of the children it's ok to go all Orwell here.

First App I'd make. Cheating spouse app. Is your partner or Ex hiding from you. Pay here for tracking info. Part of which goes to the people collating data.

Nah - go B2B not B2C. Collect all the data and sell it to insurance companies. Let the health insurance companies increase people's rates for regularly using McDonalds drivethroughs. Let the auto insurance companies bump up your rates because you are seen near racetracks or vehicle modifiers. Charge income protections and disability insures to search for claimants to see what they're doing when they think they;re not under scrutiny "Mr Smith, you claim to be unfit to return to work, but you're paying golf twice a week!"

The author is just calling some api, I don't know how much technical knowledge people here at hacker news has, but this person is talking BS. No open source software are used here.

To the author: please spend some time and try to analyze the problem first. I don't think you even get the problem right, or know what open source is. A free api IS NOT an example of OPEN SOURCE.

Not sure if trolling, but:

I'm using an open source license plate recognition system (OPENALPR), taking the produced registration number, and THEN querying the (yes, closed source) records database via web scraping. All tools used to achieve this, bar where the records are kept, are open source (openalpr, cheerio, node-horseman etc).

nope, no one on HN has any technical knowledge whatsover.... all the smart tech peeps have gone over to this thing called "myspace", you might want to give it a go :)

Reply on news.ycombinator.com