Our story earlier this week about a new smartphone app that uses AI to analyse video and enables users to report drivers suspected of speeding to the police for potential enforcement attracted a lot of comments, including from Rod King of the road safety campaign 20’s Plenty For Us, whose volunteers are trialling the app.
> New smartphone app to allow public to submit evidence of speeding drivers
Among the concerns raised about the Speedcam Anywhere app were whether police would have the desire, let alone the resources, to deal with such submissions, as well as whether there are evidential issues related to the footage captured.
However, replying to road.cc readers in the comments to the original article, King said he believes the technology “will be transformational” and that he has “every confidence that it will make a huge difference” to road safety.
He said: “I think that it’s fair to say that the forces aren’t expecting this. So don’t expect systems to automatically accommodate the new technology.
“I am confident that it will be accepted, but we do have a very patchy set of constabularies on enforcement, especially ‘where people are’. There will be pioneer forces who will see the benefits and others who are still only migrating from Gatso [speed cameras].
“The point about uploading video and report is that the report saves analysing the video. But if challenged the video can be manually analysed also. It’s all a lot more straightforward than analysing most dashcam or headcam submissions.
“It will be transformational. But some forces may take longer to transform than others. I have every confidence that it will make a huge difference.”
We’ll be contacting the app’s developers for a response to a number of concerns raised by road.cc readers, as well as getting some thoughts on it from road safety professionals.
In the meantime, 20’s Plenty For Us director King gives more detail about the app and how it functions, which will hopefully answer a number of the questions raised in the comments. He has also highlighted that more information is available on the organisation’s website.
King said:
Most speed detection ‘devices’ use a function within the device to measure the speed of a vehicle. If it’s rad or laser, it involves measuring the speed by bouncing a wave off the oncoming vehicle and measuring the doppler effect produced because the vehicle is moving.
One issue with this is that you can rarely use the device in the path of the vehicle and therefore you do not measure the vehicle speed as it is reduced by the cosine from the sight line to the direction of travel. Because the device independently assesses the speed it needs calibration.
Speedcam Anywhere is different. It is not the smartphone that measures the speed. The app buffers a video image and when you press the shutter as a car is passing and centred on the screen it selects a video snip of the previous one second and next one second. This is then uploaded to the cloud together with GPS location. The server then:
- Uses ANPR to look up the vehicle make, model and year;
- Looks up the wheelbase (WB) of the vehicle;
- Uses AI to analyse the video and locate the wheel centres;
- Finds the still in the video clip where the front wheel passes a point on the road. Takes its time stamp (T1);
- Finds the still in the video clip where the rear wheel passes the same point on the road. Takes its time stamp (T2);
- Uses the simple physics calculation that v=s/t ie v= WB/(T2-T1);
- Looks up the mapping to find the speed limit at that point;
- Creates a two second video clip overlaid with the time stamps;
- Creates an A4 report showing location, picture of vehicle, its details, location, speed and speed limit as well as time stamped stills used;
- Sends a summary back to the app.
The app user is then able to download the report and video clip for upload to the police dashcam. Here police can (if they wish) examine the video to verify the speed of the vehicle.
Because the wheels always follow the direction of the vehicle the angle of approach does not matter. You can use the app with oncoming or departing cars as long as you can see the number plate. A line of sight with one edge of the image perpendicular to the road has been found to work best.
The requirement for a Home Office Type Approved device only exists for speeding convictions. It is an anomaly that presumes that the speed is measured on the device. Where it is a video, precedents have already been set whereby drivers have been prosecuted for careless or dangerous driving based on video evidence. This includes verifying speeding.
However, this does require expert analysis and often a measurement of road markings or scenery to provide a fixed distance to measure the travel time over. Speedcam Anywhere negates the need for this expert analysis by using AI and the wheelbase of the car as a fixed distance.
Section 59 Anti-Social Driving offence only requires reasonable grounds for believing that a motor vehicle is being used on any occasion in a manner which contravenes section 3 or section 34 of the Road Traffic Act 1988 (careless and inconsiderate driving and prohibition of off-road driving) and is causing, or is likely to cause alarm, distress or annoyance to members of the public. Hence under “reasonable grounds” a Section 59 offence can be registered.
Speedcam Anywhere could gain Type Approval for the ‘process’ rather than the ‘device’. In which case police and members of the public may use it for enforcement of speeding offence directly.
This really does use a great combination of modern databases, smart analysis and AI to provide an accurate measurement of speed using a smartphone. It’s not only a great invention but also a huge step forward for enforcement. When driving in the future then, any pedestrian you see could be a Speedcam Anywhere pedestrian. Especially in urban and village settings drivers should be wary of blasting through those public places between buildings that we call streets. After all, 20 is Plenty where people are.
46 thoughts on “The smartphone app that allows public to submit evidence of speeding drivers: will it actually work?”
It would be interesting to
It would be interesting to know what the insurance companies would do if they were sent footage.
As yet ? Probably nothing
As yet ? Probably nothing because they’ve never had this before . How this app. will gain access to ANPR data isn’t mentioned though I know you’ve to be legally authorised . Interesting . What would insurers do with such clips? Their underwriting experience as far as illegal and inappropriate road behaviour goes is based on actual convictions and accidents experienced . They all realise that many if not most motorists break the law on speeding now and again , and underwriting-wise they generally don’t get overly excited for basic low end speeding . In terms of filmed inappropriate driving , again as long as an accident doesn’t occur there’s little incentive to get involved unless it’s patently wildly reckless . Jings, this could become like the film Minority Report . All that said I can see a great use in road safety campaigning by giving better quality evidence for putting in speed limits or police intervention especially say at school crossings , traffic light running (where there’s no red light camera) cycle misuse , general speeding thru rural limits and probably more . So many cars now are laden with digital technology ( gps, Speedo, radar type device for distance maintaining etc) , it can surely only be time before a ‘black box’ type device with cameras can or must be installed. If we get driverless cars , there’ll be cameras of sorts too.
It doesn’t “gain access to
It doesn’t “gain access to ANPR data”, it “uses ANPR to look up vehicle make, model and year.” This is publicly available from the DVLA. Can be used interactively via the gov.uk website (https://vehicleenquiry.service.gov.uk/) or be embedded into to applications via API (https://developer-portal.driver-vehicle-licensing.api.gov.uk/#dvla-api-developer-portal).
So in summary, the app locates and reads the plate in the image (ANPR = automatic number plate recognition), and does a lookup against DVLA data to acquire the vehicle details.
It gains ,acquires, gets – is
It gains ,acquires, gets – is all the same. I’m fully acquainted with the public available data such as Ask MID etc , wasn’t aware you could get access via the ANPR system , things move on and I’m happily retired and cycling. Vehicle details in full rather than just the reg.no. Is a good idea given the amt. of false plates that seem about . Thanks for the update .
Aberdeencyclist wrote:
I think the point is that it doesn’t need to gain, acquire, or get access to any other system. It can have its own implementation of ANPR (which is basically just to say that it identifies the area of the numberplate and performs optical character recognition on it to determine the characters). There isn’t a single, monolithic ‘ANPR system’ that it needs to use.
mdavidford wrote:
optical character recognition (OCR) is a solved problem since most corporates don’t want to actually process Invoices by hand where an application can recognise the text and be trained to know where the key data are on each suppliers documents. The AI comes in on the training to reduce the initial effort and accept new supplier or changed formats.
Number plates are a significantly easier problem given the standards that apply, though some transformation for perspective is a possible AI use case.
We weren’t talking about AI,
We weren’t talking about AI, though – we were talking about the ANPR aspect. This is just plate detection and OCR which, as you say, there are plenty of solutions for – you just need to implement one of them. The point was that you don’t need to seek privileged access to some official system.
From the description given, it sounds as though the primary use for AI that’s being suggested is in determining what part of the image is a wheel and where it’s centre is.
Yep, which is basic machine
Yep, which is basic machine learning. (spot a wheel) And is easily verifiable
Road.cc: could you ask the
Road.cc: could you ask the developers about estimated accuracy in different conditions. Plenty of variables going on here.
The arbiter of this app will
The arbiter of this app will be the courts.
If a suitably high court accepts that the app is accurate then it will be useful but you can bet Mr Loophole will pull it apart.
The police are very constrained in their use of speed camera data – they have to show it has been calibrated, operated within the operational guidelines.
On Pepipoo, drivers have still got off by vehemently claiming the camera was wrong and producing pseudoscientific justifications that bamboozle magistrates who are basically members of the public.
So, I’m not really interested in developers claims – the police and CPS won’t touch it with a bargepole because of their experience with speed cameras.
It’s different from a speed
It’s different from a speed camera. It’s calculating the speed from a video clip – IE the time taken to cover a certain distance. This can be independently verified if it was taken to court.
The trouble is that I’d like to think that if I caught/filmed a driver traveling at a speed significantly above the speed limit this would result in a prosecution regardless of whether I had the app. You can do the maths done by the app manually if you wanted. The truth is it won’t. So why would anyone think that the app will be a game changer.
I think if I downloaded the app and submitted footage I would be even more disappointed with police attitudes to road safety than I am already.
In theory yes it shouldn’t
In theory yes it shouldn’t need the app, unfortunately or predictably, it’s a complete lottery whether your local police force pursue those as routine.
The app, or the report it produces might become a game changer in that it very clearly documents the data & cause for submission, which might result in the police accepting more reports in that way.
Regardless of technology, it
Regardless of technology, it is not up to the public, the police or the CPS, as to whether this is acceptable; it is up to the courts. It would require a test case in the higher courts before you would get it accepted generally accepted. Therefore the CPS and the police are not going to be interested – not because they don’t care, but because they have been beaten up in court so many times. Take mobile phones, the law was not fit for purpose, so you have blatant use of phones where the police don’t pursue because of the loopholes – though I never understood why they didn’t simpoy use the ham sandwich approach of driving without due care.
In addition, as soon as I see someone throwing AI into the sales pitch, I’ve switched off because either:
1) it is true AI and therefore nobody really knows how it works and are going to struggle to sell it to a court as a reliable and replicatable system.
2) It is not AI, just simple maths, but the maths isn’t really simple as you need to show that the camera is still, what angle the car is moving relative to the camera (is it very far away or at a 45 degree angle?).
Anyway, the point is not that we believe the calculations are good enough, it is that a jury or magistrate believe the calculations are good enough after Mr Loophole has filled their heads with doubt. And even if it wins in one case, the varied circumstances of use mean that the next case could be thrown out because Mr Loophole comes up with a new set of doubts. The police won’t take a video to court except in serious circumstances as they would need an expert witness to prove the veracity of the case.
But yes, I submitted a video of a car obviously doing around 50mph in a 30mph, and the response was that the speed could not be verified.
This sounds promising, but it
This sounds promising, but it will take probably years to get approval and loophole lawyers will be rubbing their hands with glee.
eburtthebike wrote:
you only need HO if you want a couple of advantages, namely introducing the evidence
And how exactly is accuracy
And how exactly is accuracy ensured?
Police equipment for measuring speed and cameras rightly have to be calibrated and checked and recalibrated periodically, annually I think?
All I’m seeing is buzzwords like AI. Can you even be sure of accuracy of the recording? How does this account for a cheap phones that may likely drop frames on a recording.
Sounds like BS to me. Either an April fool or scam to get money out of investors.
Easy to count the frames and
Easy to count the frames and reject if the frame count does not match the standard FPS.
Two decades in broadcast I
Two decades in broadcast I can tell you it really isn’t easy. What’s the apps time reference for calculating speed. The internal clocks of consumer grade devices are notoriously bad, and we are talking consumer grade GPS. Then there’s the use of the vehicle wheelbase, this isn’t very large even for for something like a range rover, so what’s the margin of error? With such short variables I fail to see that the angle doesn’t matter. Then there’s the matter that you need a clear shot of who was in control of the vehicle. And again how can any of this be valid when police equipment must be calibrated to pass as evidence.
“The internal clocks of
“The internal clocks of consumer grade devices are notoriously bad, and we are talking consumer grade GPS.”
GPS is used for location so will be ok and when you say ‘notoriously bad’ what do you have to support that claim?
Fixed penalty traffic cameras don’t have a clear shot of who was driving, the letter is sent to the keeper.
GPS is also used for accurate
GPS is also used for accurate time synchronization. Every piece of IT and broadcast hardware I work with is sync’d to four NTP servers or for more critical kit PTP (precision time protocol) servers all of which sync to redundant GPS receivers. If anything loses sync you will quickly see drifting time, if it’s broadcast kit you’re liable to impact transmission with glitches and flash frames if something in the chain is out of sync with everything else. This is all kit that is a factor or tens or hundreds times more expensive than a mobile phone.
Adam Sutton wrote:
You’re describing a problem of synchronisation between different pieces of kit, which all need to keep step with each other and also with an independent reference to within fractions of a frame. I can see that is difficult.
However, if a mobile phone clock has drifted out by frame or two over a 24 hour period that makes little enough difference to the speed calculation. Moreover, a tolerance can be allowed for, so motorists filmed apparently doing 60mph in the High Street are not going to get off because a smart lawyer proves they might only have been doing 53mph.
In any case, if the app is using a GPS time signal then really we are not talking about the phone’s own clock accuracy. Unless you are saying there is a variable time delay in GPS signal processing pipeline.
No, I’m saying time accuracy
No, I’m saying time accuracy is important in determining speed. All I have done there is explain for hirsute how GPS is key to ensuring time doesn’t drift, as well as giving an example of how time is not accurate in even expensive devices reliant on time unless they are externally sync’d. You also cannot rely upon getting a GPS lock on a handheld device, particularly in a built up area like a city.
Honestly this whole thing seems pointless to me. Given most incidents are fleeting, you may as well just use your phone’s camera to record.
All of your mostly valid
All of your mostly valid points relate to speeding convictions, as the article states this apps use is targeting the section 59 offence of careless driving, thats the same bit of legislation that allows dash cam submissions for close pass prosecutions & doesn’t require Home Office type approval, calibration or any degree of specific accuracy, beyond reasonable doubt the offence was committed.
You could just as easily now, and I have done this although the result was NFA, take dash cam footage, apply a bit of Galilean physics and output the same speeding data for a careless driving submission, all the app does is send it to a computer (AI) to do that calculation bit for you. Its then down to the police force to decide what to do with it.
And the app is called
And the app is called speedcam aware, and being touted as able to calculate speed accurately. So my “claims” are pretty relevant.
“You could just as easily now, and I have done this although the result was NFA, take dash cam footage, apply a bit of Galilean physics and output the same speeding data for a careless driving submission, all the app does is send it to a computer (AI) to do that calculation bit for you. Its then down to the police force to decide what to do with it.”
And again, what is your margin of error? This is little more than throwing grass in the air to see how the wind is blowing, when applying to a vehicle breaking a speed limit unless by a huge margin.
It’s a valid point to query
It’s a valid point to query accuracy of the output data, but dont get hung up over it imo, because sec 59 prosecutions don’t require absolute accuracy, the reviewing officer just has to be satisfied an offence has happened.
For my own example the calcs I did, I think I measured over 20metres via obvious landmarks, not too dissimilar to how i calculate time gaps in road races using broadcast pictures when the OSG fails. Backed it off against my own speed sensor & distance data to validate and was comfortably at worse just 90% accurate. Which given the speed I calculated was near double the posted limit on the road I thought was satisfactory. There was no way the vehicle I reported was travelling at or close to the speed limit.
But the police NFA’d the submission because they cited the we cant prosecute speeding without a certified calibrated camera, which is absolutely correct,but they completely missed that sec 59 was an option.
And its that rather than any debate on accuracy,or type approvals, or even name of the app that’s the main issue, will the police pursue sec 59s.
Its then down to the police
Its then down to the police force to decide what to do with it
Unfortunately, that’s almost always going to be nothing. Of course, I don’t know yet that this close pass at over 50 mph by Clio PE69 OOC is going to be yet another ‘not in the public interest’, but I’m pretty sure
Thinking more about this , it
Thinking more about this , it could just be the start . Every app. created seems to spawn a rival with new enhanced capabilities that ever more powerful phones can use . Will we maybe see cycle camera systems that do the same as this app too, if linked to your phone ? Dunno whether to be pleased or worry about increased surveillance.
So if you stand on the kerb
So if you stand on the kerb holding your smartphone filming passing cars i’d imagine that you’d very quickly no longer be the owner of a smartphone in most parts of London.
How is that any different
How is that any different from making a call or checking a web page ?
No different; I’d suggest his
No different; I’d suggest his point is people have had their phone snatched out of their hand; I’ve seen video’s of just that happening in London.
I can’t believe you live in
I can’t believe you live in London if you’ve posted that.
Its good to see this
Its good to see this discussion.
The analysis report from the video clip is shown below. This explains the physics of the measurement of the speed. The wheelbase is not subject to variation by more than a couple of millimetres. The frames where the wheels pass the fixed point are shown together with their time stamp. For there to be a 5% error in the clock, and hence the time interval, the phone would be losing or gaining over an hour a day.
The clip and frames are able to be examined retrospectively in order to verify the measurement. The direction of travel of the vehicle compared to the user is not relevent because the measurement is in the same line as the direction of travel of the vehicle.
I trust this helps. Rod
Rod King wrote:
Is it possible that phones’ internal clocks really are this bad, but because they regularly resync with the network clock we never see the gross effect?
Change your settings and let
Change your settings and let us know !
Sriracha wrote:
nope – 1/100000 accuracy for the oscillators
I’d rather be safely passed
I’d rather be safely passed at 36.1 than run over at 30. Improving skills and attitudes leads to less people being run over. Improving skills and attitudes also leads to drivers being competent enough to set a safe speed, which is often below the legal limit and sometimes above. Please look up the road ahead and make sound judgements based on anticipation and hazard perception. Speed limit enforcement is one small part of road safety. Coaching road users that it’s all you need to worry about is not very helpful.
brogs wrote:
I’m not sure that first part really makes much sense even though several people like that phrase. “I’d rather have an unpleasant / frightening experience than be injured or killed” – doesn’t mean much does it?
I’m all for better training and attitude but that alone won’t make us safe of course – so enforcement is still a part of it. However:
Nope.
As a physicist what this app
As a physicist what this app proposes sounds like it will work very well and accurately to within the specified accuracy.
To those people mentioning AI: AI is only used to identify the vehicle make and the video frames where the car passes a certain point. Both of these outputs can be easily verified by human inspection.
To those people mentioning clock drifts: Quartz oscillators used in phones are accurate to within about 1 second per day, that’s accurate to within 1 part per 100’000. That’s more than enough for generating an estimated speed of travel.
To those people mentioning that the person filming could be moving: Yes, however this would inevitably result in a shaky video and blurry images unless you’re using a steady-cam, and you would have to be moving at a fast jog to make a car appear to be traveling significantly beyond the speed limit when it isn’t actually. Additionally, the photo view points won’t line up and this will be obvious.
The biggest technical hurdle is likely the image quality and frame rate of videos, but with cameras advancing as they are this is likely easily to be overcome.
The biggest operational hurdle is the time it takes to get out your phone, start the app, and hit record. By that time the car is long gone.
I would love to see this technology used by the police for speeding cameras. Set up Go-Pro sized cameras on traffic lights and Bob’s your uncle.
I changed the settings on my
I changed the settings on my phone and 8 hours later the difference to the ‘real time’ is not discernable – can only be a few milliseconds.
Your phone is connected to a
Your phone is connected to a cellular and GPS network though. It’ll be doing a periodic sync, same way when the clocks changed it updated.
This app must have Cycling
This app must have Cycling Mikey all moist in anticipation of spending his waking hours pointing his phone at passing motor vehicles..
And? I don’t race but I’m
And? I don’t race but I’m
delighted atindifferent to the thought of people being unable to sleep for excitement because Shimano are bringing out an even lighter Ultegra groupset (did I get that right)?I’m sure you have said you
I’m sure you have said you used to be a Policeman in Australia. So you are so jaded that you would just let people break the law? You do realise that Mikey lost his dad due to the actions of an illegal driver. Sounds like Autralians should be glad you are no longer in charge of Policing laws.
FWIW, from the Gruaniad today
FWIW, from the Gruaniad today:
The app cannot lead to drivers receiving speeding tickets. Since Speedcam Anywhere’s algorithm has not been vetted by the Home Office, it is not legally a speed camera, and cannot provide sufficient evidence for a police force to issue a prosecution for speeding, although the broader “dangerous driving” offence may apply if the driving is sufficiently negligent.
https://www.theguardian.com/world/2022/apr/10/speed-camera-app-developers-face-abuse-from-uk-drivers
Same article also says the
Same article also says the developers had to become anonymous due to the amount of vitriol received from irate car drivers comparing them to the Stasi…
brooksby wrote:
There’s a surprise then.
There’s a Feedback button on their website. I just sent them this: