I guess it's kind of beside the point of the article, but I wonder if the author considered using a library like Shapely[1].
At first glance, it does not seem to provide an object for dealing with a finite set of points. Using it would require first to transform the track into a continuous line, via interpolate(), then project() each waypoint to find the nearest point on that line. Issue is that this will not give me one of the original points of the track, which is what I needed.
But I may be wrong.
I've written some code to deal with this before (in my case, I wrote a distributed timer for cycle races, predicting current rider locations and gaps from sparse user reports). For me, finding nearest point on the segments of the polyline worked, but choosing only vertices would have made the time estimates terrible. Were the points somehow more fixed than that, eg towns/motels on the route?
Here is why performance mattered in my case.
I needed to predict future positions of the cyclist (see https://www.ibm.com/developerworks/community/blogs/jfp/entry...). The computation relied mostly on weather forecasts provided for each track points.
I then needed to know where to start my computation, i.e. find the track point closest to the current location of the cyclist.
One way I've approached this problem that seems pretty fast is to load a set of points into a PostGIS-enabled database, then use the `ST_Distance` function and order by distance.
[Edit]
Shapely solution:
points = MultiPoint(((0, 0), (1, 1), (4, 4)))
point = Point((3, 3))
sorted(((p, p.distance(point)) for p in points), key=lambda item: item[1])