San Diego's Connected Streetlights Taught to Recognize Bicycles (ieee.org) 24
Last year the city of San Diego installed 3,200 smart streetlights, each one monitoring 36 x 54 meters of pavement. They originally used the data to time traffic signals -- but now Slashdot reader Tekla Perry summarizes a report from IEEE Spectrum:
Developers for the City of San Diego spent months training its smart streetlights to recognize and count bicycles from just about any angle. The system is now monitoring bicycle traffic, but a few issues remain--figuring out how to distinguish between bicycles being ridden--and those doing the riding, like on a bike rack or thrown in a pickup truck.
The software has a similar problem with pedestrian-counting: When a convertible comes into view, it is counted as both a car and a pedestrian--the visible driver.
The software has a similar problem with pedestrian-counting: When a convertible comes into view, it is counted as both a car and a pedestrian--the visible driver.
Do you mean "Traffic Lights"? (Score:3)
Re: (Score:3)
Read the article
Re: (Score:2)
Re: (Score:2)
TFA only includes one photo, which appears to have been taken from a drone, and is clearly not from a streetlight or traffic light.
Anyway, if the image-recog system is misidentifying convertibles, the solution is to include more convertibles in the training set. Duh.
Re: (Score:2)
Good point about the sloppy terminology.
However, street lights also need to recognize road users in order to dim the lights when no road user is nearby. A motion sensor tuned for four-wheel cages might not give enough light to a pedestrian or cyclist, and running the lights full blast from sunset to sunrise pollutes the night sky and wastes electric energy.
Bad idea (Score:5, Insightful)
The claimed idea was to help planners decide where bike paths are needed. But they are going to put them in the exact wrong place - where people are already able to bike, rather than where people would LIKE to bike but can't.
During World Word II, the British carefully tracked where the returning aircraft were shot. Then they added armor to the places with LOWEST percent of shots.
Why? Because only the surviving planes returned. The areas totally missed where areas hard to shoot. The areas with high shots were non-essential and could survive the damage. The areas with a low number of shots meant it was possible to hit that area and if they got hit a lot, the plane would not make it back.
Similarly, the areas with high bike traffic may not need bike lanes. The areas with low bike traffic might be the exact areas where they need.
Not to mention the massive privacy invasion for minimal return.
Re: (Score:3)
You're assuming that people will look at this one piece of data that's convenient to collect and make decisions entirely on that one piece of data. That's not a bad assumption, because people are lazy. Often the smart thing to do when you get a piece of data is to use it as a guide to other data you need.
So suppose a lot of cyclists use a particular road. You should look at that road and determine whether there are accidents or other problems associated with that usage. Maybe everything's fine and you don
Re: (Score:1)
Similarly, the areas with high bike traffic may not need bike lanes. The areas with low bike traffic might be the exact areas where they need.
That depends on the goal. Just because someone can get from a to b on bicycle doesn't mean it's safe or good that they do so. I get you like the idea of increasing scope for biking, but cityplanning 101 is to segregate high use forms of transport to they don't clash in a way that slows down traffic or causes injury. If I'm already able to get somewhere by bike, but that ride exposes me to vehicles I would lobby for a dedicated path to get me where I can already go, likewise for when I'm diving 25 in a 60 zo
Preprocessing? (Score:3)
I think maybe the problem is that they are not properly pre-processing the data before feeding it into a neural network.
A car is big. A bike is not. Both are able to run at speed.
Pedestrians are small and slow.
You won't need a neural network to detect large and small shapes, and to attach velocities to them. Do that properly and you won't need a larger neural network, and you won't miscategorise too much. Also, you should make the system be future-proof for different types of bikes, motorcycles, segways and electric scooters, monowheels and whatever people may come up with.
Existing automatic systems that time traffic signals detect cars as big objects, and does not detect bicycles: Instead bicyclists can press a button in front of the traffic light to get priority. Sometimes the simple solution is the best.
BTW ... Even better would be if there were dedicated bike lanes. Biking in the midst of many cars is not fun.
Re: (Score:2)
A car is big. A bike is not. Both are able to run at speed.
This depends on how you define "at speed." The default speed limit in my city is 30 mph (48 km/h). I can sustain about half that on a bicycle.
Re: (Score:2)
virtuous use of totalitarian technology (Score:3)
Yay panopticon!
Yay ubiquitous mass surveillance!
Yay getting snooped by streetlights!
NO WAY this system will be turned to evil purposes - UNPOSSIBLE!
Ride your bike around the gulag!
Everything's gonna be okay, 'cuz Big Brother loves us all.
Where are cyberpunks when you need them (Score:2)
Who's going to spray paint cameras on these "all seeing eyes"? Where is the outrage?
Re: (Score:2)
We have bike cameras on some newer intersections in my city, and they're obvious cameras mounted to the horizontal metal post that supports the traffic lights. It is used to replace the crossing signal button for bicycles. It has a sign, "Bicycles detected by camera," or some such.
The problem is, there is no way for users to know if they've been detected. There is no way for them to predict, or easily discover, what area the cameras are successful in. This leads to bicycles remaining in the lane while waiti
Re: (Score:2)
https://i.imgur.com/AIgCBOs.png
I doubt that (Score:2)
"Last year the city of San Diego installed 3,200 smart streetlights, each one monitoring 36 x 54 meters of pavement. "
I guess they monitor 39,3701 * 59,0551 yards, as I know my imperial friends in San Diego or did they actually buy from any of the other 95% of the planet?
Re: (Score:2)
More likely the original sources said the area is 40 yards by 60 yards, and the people writing the article translated that into meters.
More People Would Ride If Streets Were Safer (Score:2)
Statistics? (Score:2)
emain--figuring out how to distinguish between bicycles being ridden--and those doing the riding, like on a bike rack or thrown in a pickup truck.
Are there really that many on bike racks and in tracks to matter?
Re: (Score:1)
It is the bike decals on the sides of buses and vans that will be fun.
3200 more surveillance cameras (Score:2)
Note to self: Don't visit San Diego, ever.
reality in San Diego (Score:2)
I'm a San Diegan, let's go over a bit where exactly these sensors are why they're there.
These sensors are in Downtown San Diego. Outside of ComicCon, this is not an area with concentrations of foot, car, and bike traffic that the current infrastructure can't serve. Downtown San Diego usually feels pretty empty, it was built for more people than currently use it. Right now, the tech and pharma industry centers near La Jolla are where the jobs are, and the cultural hotspots are the neighborhoods just north
Re: (Score:2)