Plus: The mathematical solution to catching a bike thief, how to ensure your Deliveroo doesn't come from a ghost kitchen, and a property with a very public en suite.
Binary Search is such a simple concept and various groups have been trying to get the police in Cambridge to do the same thing due to rampant bike theft, but it’s a battle. I’m not sure if it’s just institutional inertia, or just difficulty getting the message to the right people - probably both. Another challenge is the interface between the CCTV operators and the police - the requirement for a crime number and then GDPR processes seem to be used as an excuse to avoid taking immediate action, or to slow down the process so much that the videos get overwritten which is what happened when my partner’s bike was stolen in Cambridge station.
Cambridge police have become better at prosecuting bike theft, but only after years of intensive campaigning by Camcycle and other local groups - and this is in the UK’s cycling capital.
Whilst it’s easy to see why the police would rather tackle “proper” crime, I think they could gain an awful lot of public goodwill by taking bike and phone theft more seriously. It feels like such a violation when it happens and the police dismissing it by default really adds insult to injury.
Baidu and a few other Chinese firms were recently granted government incentives to trial Autonomous Electric Vehicles throughout the UAE; the aim is 25%, or 7000 vehicles by 2040. Waymo's name is absent from the big Gulf projects (Uber partnered with WeRide for Dubai), which leaves it casting about for the harder, less lucrative markets.
This would be us. Absent Gulf Incentives, infrastructure or regulatory encouragement, Waymo are likely expanding here because nobody else wanted to. The narrow, cohabited roads of our tottering metropole are not to be underestimated, and I imagine we will, no doubt, introduce their AI to the concept of frustration.
As with the Gulf initiatives, this also bodes poorly for the drivers currently slumming for Uber et al, without these bloody foreign robots coming here and taking their jobs.
Of course there is a downside to an AI taxi, and that's the membership tier. I imagine that at some point, if you don't pay the monthly stipend, Waymo will engage in the taxi equivalent of ads, and drive you past their sponsors. Pay your monthly fee, or find each trip involves a few minutes outside a branch of Dominos.
My compliments on another, excellent series of articles, and for the truly capital podcast on Prospect. Most illuminating, dear boy.
Surely the police don't literally sit there for 3 hours to review 3 hours of CCTV footage? Even without the binary search method, you can skip through it in 20 minutes.
Police have been investigating a spate of thefts of fast forward buttons from their in-house VCRs. Please inform them if you know of anyone anything found selling these cheaply on line or on the high street.
I don't think finding a grainy video of a random bloke in a city of 10 million is the slamdunk you think.
How likely are the police to identify a suspect from a CCTV video?
They would need to follow the suspect across town on multiple CCTV to his lair. Difficult if not in a car with reg plates. That's definitely something they will do for the sort of crime that gets you on crimewatch but sadly not for a bike.
Binary search overcomplicates the issue, when all major modern cloud-managed video security systems offer AI search with an LLM style interface.
Inputting something akin to "Person moving blue bike" would instantly jump to any instances across the footage (which could be days or weeks of footage) where this event took place. The days of police or security teams manually watching back hours of footage on end waiting to catch a crime taking place are long since over.
We have a Rhombus system, and used it for this exact use case over the summer with the entire process of locating the incident and isolating and sharing the clip with police taking just a few minutes.
I've ridden in / walked around / cycled around Waymos in San Francisco. Personally I found them to be far safer than an average Uber driver, though I recognise that's a low bar.
It will be interesting to see how they adapt to London's streets, which are frankly far more chaotic, and how / where they wait for riders, as London has far fewer places to wait without obstructing traffic.
Binary search can be super efficient, but it only works if the camera covers the stationary bike. If the bike is out of coverage, and the police need to find footage of someone walking past the camera with a just-stolen bike, it will probably take ages.
I followed a Wayve cab into town along the A41 a few weeks back at rush hour, and it seemed quite nervous, if a robot cab can be that. It's probably good that they err on the side of caution, but because it wasn't jumping in the gaps the human drivers were, it was noticeably getting left behind by the flow of traffic. Everything else being equal, I can't see why I'd choose a slower robot driver over a human one.
I've used Waymo in Los Angeles. Unfortunately they don't go to the airport there, but it was quite a pleasant experience, even better than driving your own car (because you're not driving). Only downside were the set pickup / dropoff points.
Thanks for the good article. There is one factual error though. It states that jaywalking is illegal in San Francisco, which has not been correct since 2023.
I saw the Waymo self-driving taxis whilst in Atlanta over the summer - if they can deal with the fast and crazy driving on the Interstate highway that runs through the middle of the city (think British Touring cars in the 90s) then they should be fine with the lower speed vehicles in London.
Cycle hire and e-scooter riders however will prove more of a challenge.
Couldn't AI be trained on the average time it takes to unlock a bike by a legit owner - ie put key in the lock and take off chain, versus the average time taken to bolt-crop (admittedly, this can presumably be just as quick as unlocking but maybe thief spends more time scanning about and prepping while standing by bike) or the time it takes to angle grind a lock off. Presuming legit means are quicker, could AI scan CCTV for images of people spending more time next to a bike and zone in on, edit or highlight only those sections of footage.
Bike thieves are extraordinarily quick, so I very much doubt there’d be any difference in the length of a person’s interaction with a bike depending on whether they’re unlocking it or stealing it. Theoretically there’s probably more value in training an AI tool to identify particular movements that are characteristic of theft, but in practice the quality of the CCTV covering most bike racks wouldn’t be anywhere near good enough for specific movements to be identified.
Binary search is an obvious thing that the police should implement, but more-sophisticated use of CCTV to identify thieves would be dependent on railway companies upgrading their cameras. Having published research on this, as well as having investigated bike thefts myself, there are quite a few problems that can get in the way of CCTV actually being useful for identifying thieves. Not least is that rail companies often don’t sufficiently consider how cameras will be used when they choose where to install them. There are many examples of cameras being installed at an angle that means they spend much of the day pointing into bright sunlight (making the thief just a silhouette) or installing a camera only to install some other equipment a few months later that obscures most of the camera’s view.
Couldn't something be trained on seeing an axel grinder? I appreciate not all thefts involve that, but might help spot reoccurrences by the same people using the same approach.
If you're only reviewing footage of bikes reported stolen then you probably don't need to worry about what kind of activity is going on - the legit owner shouldn't have collected it.
Binary Search is such a simple concept and various groups have been trying to get the police in Cambridge to do the same thing due to rampant bike theft, but it’s a battle. I’m not sure if it’s just institutional inertia, or just difficulty getting the message to the right people - probably both. Another challenge is the interface between the CCTV operators and the police - the requirement for a crime number and then GDPR processes seem to be used as an excuse to avoid taking immediate action, or to slow down the process so much that the videos get overwritten which is what happened when my partner’s bike was stolen in Cambridge station.
Cambridge police have become better at prosecuting bike theft, but only after years of intensive campaigning by Camcycle and other local groups - and this is in the UK’s cycling capital.
Whilst it’s easy to see why the police would rather tackle “proper” crime, I think they could gain an awful lot of public goodwill by taking bike and phone theft more seriously. It feels like such a violation when it happens and the police dismissing it by default really adds insult to injury.
Baidu and a few other Chinese firms were recently granted government incentives to trial Autonomous Electric Vehicles throughout the UAE; the aim is 25%, or 7000 vehicles by 2040. Waymo's name is absent from the big Gulf projects (Uber partnered with WeRide for Dubai), which leaves it casting about for the harder, less lucrative markets.
This would be us. Absent Gulf Incentives, infrastructure or regulatory encouragement, Waymo are likely expanding here because nobody else wanted to. The narrow, cohabited roads of our tottering metropole are not to be underestimated, and I imagine we will, no doubt, introduce their AI to the concept of frustration.
As with the Gulf initiatives, this also bodes poorly for the drivers currently slumming for Uber et al, without these bloody foreign robots coming here and taking their jobs.
Of course there is a downside to an AI taxi, and that's the membership tier. I imagine that at some point, if you don't pay the monthly stipend, Waymo will engage in the taxi equivalent of ads, and drive you past their sponsors. Pay your monthly fee, or find each trip involves a few minutes outside a branch of Dominos.
My compliments on another, excellent series of articles, and for the truly capital podcast on Prospect. Most illuminating, dear boy.
The crux of the police issue is that they're just not arsed, really.
Surely the police don't literally sit there for 3 hours to review 3 hours of CCTV footage? Even without the binary search method, you can skip through it in 20 minutes.
Police have been investigating a spate of thefts of fast forward buttons from their in-house VCRs. Please inform them if you know of anyone anything found selling these cheaply on line or on the high street.
I don't think finding a grainy video of a random bloke in a city of 10 million is the slamdunk you think.
How likely are the police to identify a suspect from a CCTV video?
They would need to follow the suspect across town on multiple CCTV to his lair. Difficult if not in a car with reg plates. That's definitely something they will do for the sort of crime that gets you on crimewatch but sadly not for a bike.
Binary search overcomplicates the issue, when all major modern cloud-managed video security systems offer AI search with an LLM style interface.
Inputting something akin to "Person moving blue bike" would instantly jump to any instances across the footage (which could be days or weeks of footage) where this event took place. The days of police or security teams manually watching back hours of footage on end waiting to catch a crime taking place are long since over.
We have a Rhombus system, and used it for this exact use case over the summer with the entire process of locating the incident and isolating and sharing the clip with police taking just a few minutes.
I suppose one could consider this a public nudge that removes the "but we don't have the technology" argument.
I've ridden in / walked around / cycled around Waymos in San Francisco. Personally I found them to be far safer than an average Uber driver, though I recognise that's a low bar.
It will be interesting to see how they adapt to London's streets, which are frankly far more chaotic, and how / where they wait for riders, as London has far fewer places to wait without obstructing traffic.
Binary search can be super efficient, but it only works if the camera covers the stationary bike. If the bike is out of coverage, and the police need to find footage of someone walking past the camera with a just-stolen bike, it will probably take ages.
I followed a Wayve cab into town along the A41 a few weeks back at rush hour, and it seemed quite nervous, if a robot cab can be that. It's probably good that they err on the side of caution, but because it wasn't jumping in the gaps the human drivers were, it was noticeably getting left behind by the flow of traffic. Everything else being equal, I can't see why I'd choose a slower robot driver over a human one.
I've used Waymo in Los Angeles. Unfortunately they don't go to the airport there, but it was quite a pleasant experience, even better than driving your own car (because you're not driving). Only downside were the set pickup / dropoff points.
IMHO the greatest threat to taxis and black cabs is already here and it's the lime bikes.
Thanks for the good article. There is one factual error though. It states that jaywalking is illegal in San Francisco, which has not been correct since 2023.
Reference: https://www.rmdlaw.com/blog/california-new-jaywalking-law-impact-pedestrian-accident-claims/
My one visit to SF was in 2021! Thanks for the correction, I'll put that in.
A most informative article.
I saw the Waymo self-driving taxis whilst in Atlanta over the summer - if they can deal with the fast and crazy driving on the Interstate highway that runs through the middle of the city (think British Touring cars in the 90s) then they should be fine with the lower speed vehicles in London.
Cycle hire and e-scooter riders however will prove more of a challenge.
Couldn't AI be trained on the average time it takes to unlock a bike by a legit owner - ie put key in the lock and take off chain, versus the average time taken to bolt-crop (admittedly, this can presumably be just as quick as unlocking but maybe thief spends more time scanning about and prepping while standing by bike) or the time it takes to angle grind a lock off. Presuming legit means are quicker, could AI scan CCTV for images of people spending more time next to a bike and zone in on, edit or highlight only those sections of footage.
Bike thieves are extraordinarily quick, so I very much doubt there’d be any difference in the length of a person’s interaction with a bike depending on whether they’re unlocking it or stealing it. Theoretically there’s probably more value in training an AI tool to identify particular movements that are characteristic of theft, but in practice the quality of the CCTV covering most bike racks wouldn’t be anywhere near good enough for specific movements to be identified.
Binary search is an obvious thing that the police should implement, but more-sophisticated use of CCTV to identify thieves would be dependent on railway companies upgrading their cameras. Having published research on this, as well as having investigated bike thefts myself, there are quite a few problems that can get in the way of CCTV actually being useful for identifying thieves. Not least is that rail companies often don’t sufficiently consider how cameras will be used when they choose where to install them. There are many examples of cameras being installed at an angle that means they spend much of the day pointing into bright sunlight (making the thief just a silhouette) or installing a camera only to install some other equipment a few months later that obscures most of the camera’s view.
Couldn't something be trained on seeing an axel grinder? I appreciate not all thefts involve that, but might help spot reoccurrences by the same people using the same approach.
If you're only reviewing footage of bikes reported stolen then you probably don't need to worry about what kind of activity is going on - the legit owner shouldn't have collected it.