An AI dashcam app designed to rate every driver

http://spectrum.ieee.org/cars-that-think/transportation/sensors/the-ai-dashcam-app-that-wants-to-rate-every-driver-in-the-world

by walterbell • 10 years ago

150 105 10 years ago

spectrum.ieee.org

aub3bhat• 10 years ago

Great who needs debate around static ALPR (license plate readers) data collection when we will soon have swarm of cars running with connected dashcams. The resolution is already good enough for face recognition on bystanders/people crossing street. Things will get even more interesting once we get self driving cars equipped with omnidirectional video recording capabilities.

I am very interested in what can be done, with such data. I am currently working with long duration (>20 minutes) dashcam videos uploaded by users to youtube.

Xeoncross• 10 years ago

You realize that this kind of power will be banned and put under the control of the government only right?

We can't have companies violating privacy like this. Only the government can do that.

Cyph0n• 10 years ago

But companies already analyze vast amounts of data, so I don't see how this is different.

schiffern• 10 years ago

Only the government can do that.

Why would the government want to spend the money and PR capital developing and rolling that out, when they can have a company do it for free, take the heat, chest beat about private enterprise, and then secretly NSL subpoena or crack it and distribute the intelligence to every interested agency, police department, and eventually favored corporate interest (using parallel construction as needed)?

tedmiston2• 10 years ago

Hopefully this can seed the discussion of privacy implications for tracking a driver's actions, especially tied to the (sort of) unique ID that is their vehicle's license plate.

I wonder if they'll be able to tell when someone else is driving your car from the "fingerprint" of their driving style.

btcprint• 10 years ago

There was an article recently about obd2 scanners recording information such as throttle application, brake pressure, etc and after recording information for 15 minutes could determine with X% accuracy who was driving and after 60 minutes of data, could fingerprint who was driving with 97 (or maybe it was 100%?) Accuracy.

Pretty interesting stuff

tedmiston2• 10 years ago

This is neat.

My roommate in grad school developed a somewhat related idea but for identifying someone's "typing cadence". He used it for an additional security factor in the way that a person types their passwords in a browser.

baldeagle• 10 years ago

Neat stuff, but I think that particular study had a pretty small sample size (~100). It would work from telling mom from mom or junior, but not so well at picking the zip car driver from a pool of all zip car drivers.

macintux2• 10 years ago

Today I pulled into a turn lane and stopped alongside the car ahead of me to let them know their brake lights were out. Had to honk incessantly to get their attention. Then I waited in the wrong lane (no one behind me) to pull back into my original lane to go straight through the light.

Probably would get a demerit from an AI for that maneuver.

Casseres• 10 years ago

Maybe the AI will send them an email telling them their brake light is out, so you wouldn't have to put yourself in that position.

TheSpiceIsLife2• 10 years ago

No, they'll get a fine for driving a defective vehicle and be require to make-good and attend vehicle inspection to prove. The all seeing AI enables strict enforcement.

TheSpiceIsLife2• 10 years ago

I used to do this, until one driver go out and punched me in the face. Didn't catch his plate number unfortunately, vision was too blurry.

good_dog• 10 years ago

Pulling into a turn lane, stopping alongside another car and honking incessantly is dangerous and against the law. The other driver was probably scared and trying to ignore you.

readams• 10 years ago

Man this is a really creep company. What do they imagine their business model really is? The warning thing is not a good business. Presumably they really want to sell this data to insurance companies.

dogma11382• 10 years ago

Actually this isn't a bad business model iMO they sell the data to insurance companies and you can lower your insurance costs.

And while your insurance cost might also go up if it deems you to be a reckless driver that can also have a good impact because it can force you to drive more safely in order to bring your insurance costs back down.

If they can incorporate the app with some instructional service like a weekly summary of all the misstakes you make and give you say 5 tips to drive better it can actually have quite a positive effect on road safety.

And while it's a bit "creepy" I see this to be pretty much similar to insurance companies "paying you" to quit smoking even sending you to "free" quitting smoking seminars/workshops or "paying you" to go to the Gym often even with a free yearly subscription.

Those things cost pennies to the insurance companies while saving them a lot of money in the long run, only a handful of less lung cancer cases or heart attacks a year can cover the cost of those programs for 1000's if not 10,000's of policies.

Washuu• 10 years ago

The company representative has been spamming Reddit with their videos* and advertisements as well in the various car crash related subreddits. It is getting really old seeing "NexarGuy" next to another link to a video plastered with their advertisement banner.

*By "their videos" I mean videos from users using the application that get uploaded to their servers and posted to their YouTube channel.

sokoloff2• 10 years ago

Nexar will face some ethical dilemmas. For example, should the app inform users when it spots a license plate that’s the subject of an Amber Alert?

Absolutely. I'm not one to normally fall victim to the mindless "think of the children", but Amber Alerts are rare and specific enough that they seem perfectly useful and legitimate usage of this tech. (I find most of the rest of the tech between useless and invasive, but flagging Amber Alert tags I'd support.)

noja• 10 years ago

Absolutely at the moment. That stops being okay once amber alerts start to be misused.

fidget• 10 years ago

I mean, it's been 20 years, and people seem to be using them sensibly still. I'm not convinced their misuse is a given.

sokoloff2• 10 years ago

Here's where I believe "think of the children" madness works in favor of people who think like I do. Law enforcement and the DoJ will be very reluctant to start issuing Amber Alerts for drug dealers, bank robbers, bail skippers, and the like. Even missing elderly get Silver Alerts (where they could have just as easily be lumped into Amber Alerts).

Amber Alerts for my area come to my cell phone and I can't recall getting more than one per year. Looking up stats for 2014, they ranged from 9-21 per month (nationwide). I'm pretty comfortable that surveillance over-reaches will be a long time coming to Amber Alerts (if ever).

zump• 10 years ago

Why would the average person get this app?

mk-61• 10 years ago

Let's put aside possibility to get discounts for ones perfect driving.

Think about instant posts of driving achievements to Facebook, instagram, twitter.

imaginenore• 10 years ago

These kinds of apps are inevitable. We will see apps that rate drivers, workers, and just people. I'm not as worried about the apps, but rather with someone gaming the rankings.

Both Yelp and Travelocity turned from awesome to pretty much useless within a few years. Reviews and ratings can be bought. Employees are forced to leave positive reviews.

seanp2k26• 10 years ago

Yup, they're inventing another game to play here, just like credit scores. If you know that you're playing a game and you understand the rules, you can usually figure out how to play better than most while investing the minimum amount of time playing. If you don't know that you're playing a game or you aren't aware of the rules, you'll spend a lot of time being upset and losing money, typically.

It's also true with credit cards. If you don't understand merchant fees and maximizing your rewards from purchases you'd otherwise make while simultaneously bolstering your credit score, you're leaving money on the table.

Humans will hopefully look back in a few hundred years and realize how futile and terrible of an idea it was to attempt to gamify every aspect of our lives so some company or another can extract a profit from the unsavvy, but since there is so much "value" to be "created" right now, I doubt that it'll happen any time soon under our current system of governance and beliefs about property, taxes, "free" markets, etc.

Animats• 10 years ago

I'd expected autonomous vehicles to do it, but not that it would be a business targeted at drivers. Not clear what the end user benefit is. A system that recognizes unmarked police cars would have a market, but there would probably be objections to that.

As is typical today, it uploads all the data to a server and the service operator keeps all the rights to the data. Sigh.

Using it to pass around hard braking info is not useful, though. That's what radars are good for. They'll always see the car ahead, whether it's equipped or not. The widespread use of radar-controlled auto-braking is the next big thing in auto safety. BMW, Mercedes, etc. already have it, so it's a proven technology. Not car to car communications, which don't solve any non-advertising related problem.

lpage2• 10 years ago

Assuming that insurance is correctly priced, buying a policy in any form is negative expected value. The companies that write such policies would slowly bleed dry if that wasn't the case. However, that's irrelevant unless you have the bankroll to withstand tail risk. Given that most people don't, as a society, you pretty much have to make things like auto liability insurance compulsory.

If you're a "good driver," as defined by a lack of tickets and accidents, you get discounts. There's really no way to tell if you're _actually_ a "good driver" because "good driver" is a hidden variable, and the only observables from an insurance standpoint are tickets or accidents. I know horrible drivers who never have accidents and avoid tickets via luck or skill. Conversely, you can be a good driver who's unlucky enough to have multiple accidents in a short period of time. Outcomes = baseline truth + random noise.

Most corporate data grabs bad news, but I'm ok with it when it comes to pricing insurance (outside of health). If I run a business that has rock solid, well enforced policies in place to mitigate risk, I don't want the insurance company's priors over my company's sector driving the premium. I only want to pay for coverage on idiosyncratic risk - the risk that remains after addressing everything that I can. Same goes for driving habits. Things like health are almost entirely idiosyncratic risk (ergo, let's agree that doing this for health is a bad thing), but when that's not the case pricing policies using true states (the baseline) and not observables (baseline + noise) puts you as the policy buyer more in control, not less.

That said, I have major concerns over the implementation. A breach is all but inevitable, so there needs to be a means of updating models online, and rules against data retention. It's also bad news if flawed models or corner cases incorrectly price someone out of a policy and there's no sanity check. Unfortunately, that's not too different from the actuarial status quo.

Lxr• 10 years ago

Assuming that insurance is correctly priced, buying a policy in any form is negative expected value

Actually it's quite rare for insurance companies to record an underwriting profit, I believe. Their profit generally comes from investing the pool of premiums. If you invested your premium instead of buying insurance however, then I suppose the expected return would have to be better.

lpage2• 10 years ago

Yes that's totally correct but they're factoring the entire reinsurance/reinvestment pipeline into how aggressively they can price the policy in the first place.

FeepingCreature• 10 years ago

negative expected value

This would only be true if the value of money was linear.

The value of money is not linear.

A 10% chance of a 1000$ loss can be far more harmful than a 100% chance of a 100$ loss, especially at low income.

kwikiel• 10 years ago

That's called expected utility.

Most people on HN knew about expected value but they don't know too much about expected utility. I suggested reading about it and St. Petersburg paradox

AndrewKemendo• 10 years ago

This looks like an excellent way to build the most robust car tracking eco-system ever.

quietplatypus• 10 years ago

I'm all for this. Especially in Bay Area traffic. It looks like about 95% of the time I get some tailgater because I leave enough room in front of my car to brake. If I dont' want to get tailgated, I end up tailgating. I seriously wish these people would either shape up or stop driving cars altogether.

Qantourisc• 10 years ago

Small issue to potentially give up privacy for. This is rather a job for the police and sensibilisation campaigns.

Also think of this way: if you both tailgate, and you have to brake, you will be hitting both the one in front of you, and getting hit in the back. If you don't tailgate and only get tailgated, you will only get hit in the back. (And have a way better defence for your insurance company.)

(PS I'm going to assume this is tailgating on the highway ?)

nl2• 10 years ago

I've considered building something like this for bikes.

For those who aren't cyclists, there are already informal networks of people who share helmet-cam videos of bad driving behaviour, with people who attempt to read license plates etc

cgriswald2• 10 years ago

In my experience as both a cyclist and a driver in the Bay Area, I have no interest in the opinions of other cyclists on driver behavior. Cyclists largely disregard the rules of the road and then blame drivers.

As a driver, I have been at the receiving end of abusive gestures and language even when following the law. In one case, I yielded to a bicyclist who was in blatant violation of the law[0], because failure to do so would have endangered him. He showed me his middle finger. Another was angry that I was in the bicycle lane making a right-hand turn, as required by law and common sense.

As a cyclist, I have given up trying to educate my fellow cyclists, and generally won't ride with others anymore.

[0]I was already turning through the intersection after having made a stop. The bicyclist arrived in the middle of my turn and entered the intersection without stopping for the stop sign.

Edit: Formatting.

nl2• 10 years ago

So.. cameras == good for both, then?

a_imho3• 10 years ago

As a cyclist, I have given up trying to educate my fellow cyclists, and generally won't ride with others anymore.

I don't know about the bay area, but in my city you can't really group ride, bike roads are single lanes. Yet people ride side by side all the time.

Personally, I found the best way to change the culture is to show an example. There is clearly an ongoing bicycle boom, and it shows many just follow the customs e.g. dashing through reds is acceptable (which is debatable) but it quickly turns out stopping for 20 seconds is not that big of a problem. It is quite interesting though, the same people would never do that in a car.

bloaf• 10 years ago

This strikes me as one of those great and terrible ideas. It has a tremendous potential for good, but it is one opt-out checkbox (Share the data you collect with the police to help catch criminals!) away from convincing the surveilled to create their own big brother.

http://asofterworld.com/index.php?id=346

qmr• 10 years ago

I highly doubt there will be such a checkbox. The "permission" to share will be buried in a EULA no one will ever read.

seanp2k26• 10 years ago

And every company who will sell you the compulsory insurance will have that verbiage in their EULA, thus making the choice "agree" vs "can't drive".

dclowd9901• 10 years ago

There seems to be a disturbing trend of Israeli developers creating software that crosses some serious ethical lines. A few offhand: this, an iPhone backdoor exploit they're hiding and have provided to the us government, I believe the biggest license plate reader developer is Israeli.

Can anyone speak to this? Am I observing this on some kind of bias?

dublinben• 10 years ago

This is probably just confirmation bias on your part. That being said, based on the political situation in Israel and the security culture in that country, it isn't surprising that many tech companies there would be developing tools that perpetuate that status quo. As a contrast, you would probably never see an app like this coming out of a hackathon in Gaza.[0]

[0]https://news.ycombinator.com/item?id=11858963

chillingeffect3• 10 years ago

It's not just you. With a heritage of Soviet-era immigrants and an atmosphere of paranoia, the space of "technological solutions" in Israel looks different than it does in other cultures. If I had a moment (in a meeting atm), I'd go through and research some of the strides their scientists have been making in oppressive technology.

Also, Stephen Hawking himself has recently participated in a boycott of Israel: [1]

[1] http://www.cbc.ca/news/technology/is-stephen-hawking-s-acade...

syphilis2• 10 years ago

I wonder how well this will scale with time. I've certainly done things while driving that were reckless to varying degrees, I suspect most drivers have as well. How long until 50% of cars on the road have been tagged doing something reckless? Of course I can imagine a number of ways to handle this.

On a more broad note, I wonder how useful it really is to know that a specific vehicle acted recklessly sometime in the past. Seeing someone behaving poorly is good reason to avoid them on the roadway. However I also imagine a scenario where one driver did something to warrant an alert, and some time later every car on the roadway is trying to avoid that vehicle. Is there really any danger from that vehicle in the current situation?

What I would like to see is a map showing where the most incidents occur, and what type of incidents occur there. I assume most drivers genuinely set out intending to behave well, but do not for various reasons. Identifying locations where incidents happen would give civil engineers some valuable data about how their designs impact behavior, and would give police valuable information about where their presence would be useful.

Bombthecat• 10 years ago

I will go even that almost 100% will be tagged soon or later.

What we should be interested in is the top one percent.

mcbrettsy2• 10 years ago

So something like http://comroads.com ?

Angostura• 10 years ago

I presume that the system would simply indicate the proportion of times that a vehicle is seen driving and performing a dodgy manoeuvre. If I'm seen on the roads 1,000 times and seen performing a dodgy lane change once (and I would like to apologise to the Mondeo driver on the A21 yesterday), that's the fact that should be recorded.

amelius• 10 years ago

Indeed, a proportion tells us something about the probability of making a driving mistake. A boolean tells us little in comparison.

seanp2k26• 10 years ago

The thing is that there's not a 1:1 relationship between a "risky" maneuver and causing an accident. An example could be an F1 driver who takes curves at 20mph over the recommended speed in a sports car. Compare this mentally to a Prius driver who drives under the speed limit in the left lane. The Prius driver is "being safe" but causing a long line of tailgaters and people passing unsafely on the right, as well as agitating people and causing them to drive more aggressively. I doubt that the first few iterations of this system will take psychological effects on other drivers into account (if they ever consider that / have a profit motive to do so at all).

AJ007• 10 years ago

There is definitely a gradient of severity with "reckless" driving.

Have you ever driven full speed through a red light? Sat at a complete stop of a freeway in an attempt to merge in to traffic? Taken a U-turn on a freeway? There are things that don't cause an accident, but once a driver does, they should probably not be allowed behind the wheel for at least a year, some things possibly even a lifetime ban or some additional evaluation.

There is an interesting dichotomy because smart phone usage has really caused a big problem with distracted driving. Even Geico experienced an increase in claims last year, and I suspect this was why. When I watch traffic sometimes over 50% of drivers are looking at a phone screen rather than the road. At the same time, mobile sensors are enabling something which will end with roads being extremely safe and killing very small numbers of humans a year.

There is a lot of talk about privacy and freedom on the internet. Privacy and freedom on the road is dead in all dimensions. You won't be anonymous, you won't be able to take go somewhere when the police/authorities are looking for you, there may not even be private ownership of any vehicles in the US in 10-15 years.

randcraw• 10 years ago

Why won't the app report every tiny detail on the driver it knows best... You? This seems like a sure fire way to invite yourself to higher insurance rates and a big presence in every law enforcement database that tracks peoples' movements.

There'd have to be a very big reward for me voluntarily becoming Big Brother, and especially for tattling on myself.

kylec• 10 years ago

The real purpose of this app isn't the rating system, that's just the hook to get people to run this app and point the camera out their dashboards. The real purpose is to crowdsource collecting data about the drivers on the road in order to sell that data to interested parties.

frgewut2• 10 years ago

I can imagine a lot of applications for these types of systems: identifying undercover traffic police cars, finding whether someone is following you, identifying friend's cars (similar to FB's "Share your location")

People would love such systems (and I'm afraid our privacy is gone for good).

golergka2• 10 years ago

This debate happens on HN again and again, but still: what privacy are you talking about in public space? Nobody has a problem with one person with a dashcam — why suddenly connecting these dashcams to a network that shares information is a problem?

JulianMorrison4• 10 years ago

Meh, all the purposes are real. It's a win-win. Get warned of idiots, collect data that will be useful to insurers (and probably to self driving cars too, I can imagine Google cars giving detected hotheads a wide berth).

I can see one group of people having a legit beef with this though: car rental companies. One idiot behind the wheel, and their fleet car might as well go straight to scrap.

ucaetano• 10 years ago

"One idiot behind the wheel, and their fleet car might as well go straight to scrap."

Unless, of course, they pay this nice fee to have their fleet cars whitelisted in the app, call it "bad driver protection money"!

dsfyu404ed• 10 years ago

Maybe there'll be an opportunity to drill down in the data. There's a huge wide spectrum of people who suck at driving (senile old people, oblivious soccer mom, boy-racer, etc) and I would be very surprised if they're all roughly equivalent.

chillingeffect3• 10 years ago

That's imagining this system in the most perfect light: shiny, new and not gamed.

The opportunities for misinformation and asymmetricality are huge. Don't like someone? Clone their license plate for a day and drive like an ass. How could you ever prove your innocence?

Loan your car to a friend and they could tarnish your reputation. It would make it inhibitive to share vehicles.

This is technology that drives us farther apart as people in favor of a few insurance companies making even more profit than they do already. Think critically.

It would also give police even more excuses to profile innocent people.

If you want people to drive more safely, then think up ways to get more people to drive more safely. Don't expand the surveillance mechanism to every moving vehicle.

a_imho3• 10 years ago

It really sounds like a slapped on excuse for the next incarnation of the data collection biz. How would they generate revenue if not by selling the collected data to 3rd parties? Because imo that is the most straightforward option. Missing this bit entirely in the article is disingenuous and concerning. I'm not too sure about opaque code rating drivers either, though people might be ok with it.

FilterSweep2• 10 years ago

agreed - it is a huge sell for insurance companies.

electriclove• 10 years ago

This type of system will inevitably become commonplace and through abuse by insurance companies, it will push people to adopt autonomous vehicles.

peylix2• 10 years ago

I read somewhere that the app doesn't only rate the drivers around you, but also looks at the person running the app and rates his/her driving. Add the inner facing camera recording they're using and why anyone would want to actively participate in something like this is beyond me.

frgewut2• 10 years ago

When these types of tracking become common, government tracking will look like a child's play.

hmate9• 10 years ago

In a couple of years it would just be driverless cars rating other driverless cars.

Google sees the car in front is a Tesla: 1 star :)

tychuz• 10 years ago

Car is a very crappy identifier. I've bought my car used, so at least one other irrelevant person (that is NOT ME) drove it. It also gets driven by my girlfriend and on two occasions by my two non drinking friends.

tonylemesmer• 10 years ago

In the UK a few years ago there were reports of people's license plate numbers being duplicated and stuck onto vehicles that matched the genuine vehicle's description and then used for various crimes, like filling up with petrol and driving off without paying.

This tech would be susceptible in a similar manner - stolen vehicle goes out and drives badly etc. etc. The only identifier would be the licence plate. How about using their phone account as the identifier rather than the vehicle?

Anyway sounds like an idea ripe for misuse and "feature creep".

drdeca2• 10 years ago

An idea to address privacy problems with this:

1) Only have the information for licence plate codes of people who have opted in be stored in a way associated with the specific licence plate code.

2) Have the system encourage opting in as part of setting up the system in one's own car.

3) Have an id which is used for all cars which have not opted in, so that cars which have not opted in have a collective rating. Possibly split this up by general location, and maybe time of day.

4) Because most of the cars in a given area are there often (or, at least something similar to this is true) , it may be worthwhile for, for the cars that do not opt in, to still record behavior of those cars, but to not send it to a central location, only keeping it within one's own car. Cars which have not been seen in a long time can be automatically removed from the local storage. This should not significantly harm privacy, because it is basically the same as remembering that you saw that car before and that it tended to be driven poorly.

5) Possibly (this might always be pointless/counterproductive. I'm not sure.) , in some cases only store a hash of the license plate codes, instead of the license plate code itself?

A) I think this variation should be sufficient to protect privacy, because it only does widespread collection about identifiable individuals if they they consent to it. The information it collects about individuals that did not specifically (and presumably intentionally) express consent is only collected in aggregate about individuals in a geographic region, which does not personally identify anyone, or only collected by individuals and not shared, which is not much more of a privacy violation than an individual with excellent memory looking around.

B) So, while this variation might be less effective, I think it should be sufficient to protect privacy.

lamontcg• 10 years ago

I should probably be horrified by the privacy implications, but I'm really not. I'm already trying to tone down some of the headgames that I've learned to play while driving. This would help by allowing me to just let everything slide and upload the video tagging data later.

jgalt212• 10 years ago

Amen to this. There are so many horrendous and unsafe drivers on the road. Their actions must be curtailed.

An AI dashcam app designed to rate every driver

Related Stories