[oh_sigh]

Problem is you can't tell your speed in a tunnel. I propose training a ML model to be able to estimate the distance a camera moved between two input images. You could probably use google street view as training data. Then, instead of relying on fancy satellites, its just a few billion numerical calculations.

[Someone]

I would think Google street view data of tunnels would be a challenge to train on. I’m not aware of standards for distance between lights, and apart from that, they tend to be fairly dark. Also, overtaking often is forbidden inside tunnels, so you can’t easily use the size of road markers, either.

Even if that works, doing sensor fusion with a gyroscope and an accelerometer probably is both easier and cheaper.

[gleslie]

Yeah - I put this together in an afternoon.

[gsich]

There are chips for that (dead reckoning).

[lolc]

Withoug a speed sensor, dead reckoning is really hard.