I wonder if a slim local #AI model (not an LLM, obviously) could be trained to automatically ingest dashcam footage and generate useful data from it like appropriate GPX tracks, but also to feed a UI that recommends #OSM updates based in detected image data for the footage owner to quickly flip through and validate before upload. Anything well visible from the street or bike path could be very efficiently updated this way.

Would require really skilled devs… but I think the idea is neat.