3 ms·

I wonder what the AI part is? It seems you could achieve the same thing with a motion triggered or beam-tripped camera. If so, why use "AI" for this?

by giantg2 6mo ago

sowbug 6mo ago
Helps justify the $400/bus monthly "technology fee."
VladVladikoff 6mo ago
There are models that do license plate recognition, ie to find the specific box in an image that contains the license plate. Then there are also OCR models to read the characters from the plate itself. Both have been around for a long time and are not particularly revolutionary technology.
- giantg2 6mo ago
  That's more like machine vision rather than real AI. But you dont even need that when the video is being sent to a human anyways.
  - rtkwe 6mo ago
    They'd be at least minorly incentivized to cut down on false positives because 1) they had a review step before sending to the police and 2) if they're cheaping out on that first review and sending loads of false positives to the cops for citations they're likely to lose their contracts.
  - VladVladikoff 6mo ago
    Are LLMs “real AI”? What exactly is “real AI”?
- a_random_name 6mo ago
  The post office has used OCR machines (optical character recognition) since the 80s. I think the 80s machines used something more like hand built digital image processing to find the characters, but OCR is absolutely not new.
johanvts 6mo ago
AI just means software at this point.
rtkwe 6mo ago
It's the AI term bloat where any application of machine learning or vision becomes AI, it's been happening for 2+ decades now I think it's pretty well established and here to stay. That said it does need to be more complex because to produce as few false positives as it can, which would cost BusPatrol money to review in their first pass before sending to police (so there's at least a minor incentive to reduce them), they would have to determine where the car is and if it's required to stop not just trigger if a car passes by while the arm is out. Laws vary a lot by state but usually if there's any kind of real median traffic in the opposite direction is not required to stop so it would at least need to detect if that is present (or work off a database that knows where all the divided roads are in the area I'm not sure which would be cheaper but mapping feels harder and having the camera able to determine if there's a median can be deployed anywhere while mapping data is location specific).

I wonder what the AI part is? It seems you could achieve the same thing with a motion triggered or beam-tripped camera. If so, why use "AI" for this?

by giantg2 6mo ago

I wonder what the AI part is? It seems you could achieve the same thing with a motion triggered or beam-tripped camera. If so, why use "AI" for this?

sowbug 6mo ago

Helps justify the $400/bus monthly "technology fee."

VladVladikoff 6mo ago

There are models that do license plate recognition, ie to find the specific box in an image that contains the license plate. Then there are also OCR models to read the characters from the plate itself. Both have been around for a long time and are not particularly revolutionary technology.

giantg2 6mo ago
That's more like machine vision rather than real AI. But you dont even need that when the video is being sent to a human anyways.
- rtkwe 6mo ago
  They'd be at least minorly incentivized to cut down on false positives because 1) they had a review step before sending to the police and 2) if they're cheaping out on that first review and sending loads of false positives to the cops for citations they're likely to lose their contracts.
- VladVladikoff 6mo ago
  Are LLMs “real AI”? What exactly is “real AI”?
a_random_name 6mo ago
The post office has used OCR machines (optical character recognition) since the 80s. I think the 80s machines used something more like hand built digital image processing to find the characters, but OCR is absolutely not new.

johanvts 6mo ago

AI just means software at this point.

rtkwe 6mo ago

It's the AI term bloat where any application of machine learning or vision becomes AI, it's been happening for 2+ decades now I think it's pretty well established and here to stay. That said it does need to be more complex because to produce as few false positives as it can, which would cost BusPatrol money to review in their first pass before sending to police (so there's at least a minor incentive to reduce them), they would have to determine where the car is and if it's required to stop not just trigger if a car passes by while the arm is out. Laws vary a lot by state but usually if there's any kind of real median traffic in the opposite direction is not required to stop so it would at least need to detect if that is present (or work off a database that knows where all the divided roads are in the area I'm not sure which would be cheaper but mapping feels harder and having the camera able to determine if there's a median can be deployed anywhere while mapping data is location specific).