Implement Kalman filter about movingpandas HOT 6 CLOSED

I was going to look up a dataset we played with a while back and you just reminded me that it was called geolife. That's certainly something we can start with.

Constraints can be soft but starting with none seems like a plan. We'll take a look and do things in a way where upgrading to more sophisticated algorithms will be easy later.

from movingpandas.

There's another project, Stone Soup (https://github.com/dstl/Stone-Soup) which aims to provide an open-source repo of "tracking" algorithms (including, but far from limited to, Kalman filters) and has recently been improved to make use of geospatial data structures similar to those that I think are part of movingpandas. Can you specify where any Kalman filter or related algorithms would sensibly sit and/or what the input/outputs would sensibly be?

from movingpandas.

Thank you for the pointer to Stone soup, Simon! The main reason for adding a Kalman filter is as a way to remove outliers from (or smooth jittery) GPS trajectories, It would be conceptually similar to the already existing TrajectoryGeneralizers: input is the raw trajectory, output is the cleaned trajectory.

from movingpandas.

Am i right to think that the outliers are uninformative? If the task is outlier removal for each sequence of reports, we probably want something a little more sophisticated than a kalman filter (eg an IMM, which is also in Stone Soup already). If we want to use other information to help (eg we know the trajectories are on a known road network), we could opt for something more sophisticated still (eg a particle filter). Are there specific use cases or example test data that exemplify the issue?

from movingpandas.

Outliers can be considered uninformative, often stemming from GPS errors or data transmission errors. So far, MovingPandas deals with unconstrained movement data (i.e. not network-constrained) so I'd prefer to start with methods that don't require additional data beyond the movement itself. I haven't compiled any test data for this usecase yet. Does Stone soup include some that we could use?

from movingpandas.

In the Geolife sample, we can find quite some noise: https://mybinder.org/v2/gh/anitagraser/movingpandas/binder-tag?filepath=tutorials/tutorial_getting_started.ipynb

from movingpandas.