Re:Google will delete new accounts' data by default in 18 mo...
by notagoodidea on Thursday June 25, 2020 @05:45PM
(#23646129)
One way to do it will be to get OpenStreetMap at least the road and path database. The quality of it vary at lot by cities/countries. The easy way, is to set a reasonable buffer for you GPX points (dunno, let's say 10m at least) and check if they intersect with the roads/path/street they connect to. It is a bit like your way of filtering (and probably a bit like Google must do it : Check around infrastructure and filter based on the probability to be on that road/path/poi between the two points before and after.)
If you want to go the extra miles, I remember this talk of Ilya Zverev "Hundred thousand rides a day"[1] seen at FOSDEM 2019.
[1] https://archive.fosdem.org/2019/schedule/event/geo_gpxtraces...
One thing I've noticed is that Google must obviously do a lot of filtering and correction to the data, because my data is nowhere as clean as I've seen on Gmaps, and I haven't found a way to replicate that in any way. My data is totally sufficient for whenever I've needed it, but it's not perfect. The most I'll do is filter out data that is ridiculously out of bounds from what the data directly before and after say, but I wish there were a better solution (and if there is, I haven't found it yet).
I think one big problem with Google collecting all this data is that you don't own it. You can't choose to use this location data to solve your problems in a way that you'd prefer. Instead, you have to hope that Google has covered your needs for the data and if not, tough luck, you're stuck doing a lot of manual work.