I am currently building a small application that visualizes LatLng-Paths over time. Basically, what I have are lists of WSG84-LatLng-coordinates, all of them with a timestamp. The application should display them moving over time, so I have to interpolate the position between LatLng with time A and LatLng with time C at time B. I have millions of these paths and the position has to be calculated in real-time, so efficiency matters. This is all standard stuff, if you are in a cartesian x,y-system, and at the moment, I am treating the LatLng-coordinates as cartesian and do a simple linear interpolation.
However, I am aware that LatLng-coordinates are not cartesian coordinates and that linear interpolation can be problematic on the larger scale because the earth is, of course, not a plane. I dont expect to run into any problems here because the distance between two coordinates at the moment is (at the most!) one kilometer, but I want the “get it right” at this stage of the development process, which means: i want the app to be able to interpolate correctly on a larger scale.
Now, my idea is to project the whole dataset to the plane using spherical mercator projection and to do the interpolations on this x,y-plane. Is this a reasonable approach? Do I win anything doing it that way?