AI cam: 5,000 sidewalk riding reports to NYC 311 in 60 days: e-bikes and bikes belong in the streets

Описание: This video is 22 hours of footage submitted to New York's 311 system. I sped it up 8x to fit on YouTube as well as allow for faster watching. Please note that these reports are for people riding on both sides of the street so look closely and you'll see bikes on both sidewalks.

Background:

E-bikes, mopeds and bikes ride on the sidewalk in New York City constantly even though it's it's dangerous and illegal. It scares people every day.

It was always annoying us that vehicles would whiz by us on the sidewalk and swerve at the last minute to avoid hitting our dog, but when an e-bike almost hit my wife in November, I had enough.

I setup a camera in my apartment window and used AI to detect vehicles on the sidewalk, and wrote code to submit those clips to 311. However, I manually review every single clip to verify it's authenticity prior to submitting.

Some people think that it's harmless to ride on sidewalks and claim cars make it too dangerous to ride on the streets. That logic trades rider safety for pedestrian safety. People are continually being hit with many of those injuries going unreported. I support adding bike lanes to every street in the city, but it doesn't mean people should break the law if they are not there.

The NYPD has been in communication with me and have been very professional and I appreciate that they have many more serious crimes to address in New York City.

I have given them all the data as well as a graphs that predict at what times the infractions will occur. No where else in the city can you predict with this much accuracy the probability of a law being broken. Given the overwhelming evidence that this a problem, they should be taking more action to get riders off the sidewalks.

After I submitted 800 reports I started seeing squad cars stop by for small periods of time and tell people to get off the sidewalk, but many times bikers just ride past them.

Neither precinct has responded to my inquiry about how many tickets have been written. I fear that the answer may be near zero.

Councilwoman Carlina Rivera's office could care less and blew me off when I tried to press them about what legislation she will support or introduce to crack down on violations of sidewalk safety. All she will tell you to do is "report it to 311." Well, that's what I did.

Method and Technology:

I set up a camera and pointed it at the street. I then fed it through Frigate running on a mini-pc with a Coral AI Accelerator. I set Frigate to detect two wheeled vehicles in two zones I drew for the sidewalks. Frigate then exposes these events with an API that I access on a node.js server I wrote. Frigate generates a lot of false positives, so every clip must be manually verified which is a massive time sink, but completely necessary as the NYPD will show up at your door the second you mis-click and claim "false reports." (Yes, this happened to me even though I had submitted 2,000 311 requests, and only screwed up a handful of times.) So... I set up a web UI that forces double verification of every clip from Frigate. Finally, it will submit 311's for all pending detections through puppeteer browser automation.

This was working fine but I am generating around 80-100 service requests per day. Unfortunately, for every submission I had to reject 2 to 3 bad detections from Frigate. It was exhausting. So now, after Frigate generates events, I have a second, much much stronger AI (YOLO V8 via Ultralytics in Docker) that analyzes pending Frigate events every 5 minutes on a PC with an NVIDIA RTX 4090. This greatly reduced false positives. Note - I also had to seriously upgrade my camera in order to get night detections and capture the faster movers.

The music is from :
https://musopen.org/music/14910-the-f...
https://musopen.org/music/performer/t...
https://musopen.org/music/download/13...