Comment by uuyi | Hacker News Reader

Comment by uuyi | Hacker News Reader

Parent Full thread

Lot of assumptions here on measurements which really make this laughable. You need the following to make the measurement have any sort of validity:

Assuming simple s = d/t

1. The wheelbase data source is accurate. (d)

2. The vehicle type identification is accurate (d)

3. The centre of the wheels is identified correctly (d)

4. The wheels being on the same vehicle are identified correctly (d)

5. The measurement resolution is limited by pixel resolution so the image needs to be large enough to not lead to quantization errors (d + t)

6. The direction of travel of the vehicle is accurately determined (rate of motion is different if car moving away) (d)

7. The rotation of the vehicle is accurately measured (wheelbase compressed if turning away from the camera) (d)

8. The phone's camera is digitizing frames at the same rate without jitter (t)

9. The video source is actually coming from an actual camera and has not been modified by the user to incriminate them (d+t)

10. The camera isn't being moved in sympathy with the reference frame (d+t)

11. The reference frame is identified correctly (d+t)

12. the correct set of wheels on the vehicle are identified (d)

13. The shape of the lens of the camera (wide angle lens distortion) (d+t - possible calculus fun there)

I'd take the other arguments apart but I don't think I need to.

Edit: keep finding more problems with this.

Don't forget, a defendant has the right to subpoena the device that took the video/measurement. Users should be aware that they can essentially lose control of their phone during this process.

There's a reason most police departments forbid personal devices from being used to collect evidence in most circumstances.

And potentially a right to subpoena the source code to the software in some circumstances I believe.

(6) and (7) are not assumptions - the angle you are looking from doesn't make a difference, the time taken between each wheel crossing the same part of the image is the same.

What happens if the plate is incorrectly read? I have had this with my old vehicle leading to being fined for entering a car park but not leaving it.

I disagree with points 6 and 7 entirely. You can observe wheelbase compression standing on a tight bend with your eyes which are fairly low precision.

Also throw an extreme scenario on the table. If you're travelling at 70mph, that is 31m/s or 1m image resolution for a 30fps video. If your wheelbase is 3m that's a resolution of three frames to pass which is well below any threshold where a credible measurement is possible as you convert from discrete to continuous time.

Of course that brings in the problem that lower speed resolution is probably better so I should probably be driving around at 70mph in a 30mph in my 6x6 Land Rover 130 (sorry couldn't resist that).

I think there are too many assumptions to make a reliable measurement within a tolerance for the purpose described. The existing measurement devices have to be calibrated and must be operated within certain constraints to make an accurate measurement otherwise the information is inadmissible. Your app doesn't contain 1/10th of the assurance or competency that would be required. All it does is allow really annoying people to annoy people.

Reply on news.ycombinator.com