Using a few manual steps to cut corners I now coded the conversion routine to generate a .GPX file (for a tracklog) from the Google Earth .KML file of path tracing (manually merged segments). The process is a bit crude but only has to be done once, so what.
Here’s a few of the major discrepencies:
These are probably due to the low resolution of the tracklogging when faced with rapid curves (tracklogs points are taken about every 5 seconds).
And another (where trail has to exit onto footbridge along Center to cross creek, then loops back under (bridge may cause GPS to drift):
And another, where we may have two problems: 1) GPSr may lose track going under I-80 since it is shielded from satellites, and, 2) tracing on Google Earth of the exact path of trail is hard since it can’t be seen due to I-80 covering it:
And another, again sharp curves get poor tracklogs due to slow recording and relatively fast speed (tracklogs are from bike, to get better accuracy I should walk these):
And finally, where the Big Papio ends and we go on car bridge to connect to the Keystone on other side of creek:
The few bits of tracklog near the parking lot are due to a ride where I parked and then walked bike up to trail but was recording the track. There are a number of “extra” bits of tracklog in my cumulative set of logs that require some manual editing to clean up so they don’t confuse the trail location.
So, in general my tracklogs correspond fairly closely with what I hand digitized from Google Earth, but it still remains as an interesting piece of algorithm design how I’d average multiple tracklogs, in various segments and directions, into a single “average” trace of the trail, esp. in those areas where I have anomalous data. In theory Google is doing this for plotting roads in areas with few maps with no pre-knowledge of where the roads are and/or without manual intervention – quite a trick that I’d like to duplicate. Of course they may have a lot more data but I can probably simulate (for small areas) getting as much data as they have.
So on to more algorithm research, again not to find a well-known trail, but to develop techniques for mapping uncharted trails.