Stop Sign Detector
So there's this Stop Sign
Everyone Runs this Sign
My house has a good view of it from the back...
But of course, I can't spend time counting cars...
But a webcam attached to a Raspberry Pi sure can
But I *can* get my housemate to drive his car around the block until I detect it
And then I can really smooth out my object detector code until I get crisp cars every time...
...and *then* I can capture summary statistics about the motion of objects to determine likely violations, like driving backwards
📊 Stats (I moved so results are now static)⏱️
Total hourly stop sign runs
Total daily stop sign runs
(Note that these totals result in separate totals due to the difference in resolution of hourly vs. daily)
3,354 Total scans, 10,024 total suspected cases
207+ Hours collected, spanning 53 days
≈6,360 total cars seen, ≈4,223 seen long enough to measure. An estimated ≈39,569 cars have passed the sign, with an average of ≈736 per day.
Machine learning on stop violations is still pending - too few actual stops to accurately assess. As of now, 90.27% of all stops are violations according to human assessment, leading to an estimated 5,741 observed violations, with an estimated ≈35,719 total stop sign runs since monitoring began.
🏋️ Help Train the Machine 🏋️
(Note that machine learners are now shut off - pages are left for archival purposes only)
Train Car Presence
11,768 total votes recordedCurrent Machine Accuracy: 92.13% (see examples)
Train Full Scene Check
3,805 total votes recordedCurrent Machine Accuracy: 88.83% (see examples)
Train Stop Violations
3,844 total votes recordedCurrent Machine Accuracy: 93.42% (see examples)