I need a way to index and retrieve a large amount of track data. To do this I need to derive from every track as many attributes as I can. For now, I derive

  • Name
  • Tortuosity
  • Speed (average, max)
  • Covered Distance
  • Length (in time)
  • Bound (min lat/long and max lat/long)
  • Trip mode (car, bus, plane)
  • Presence of tunnel (estimated length)

But I need extra ideas, spread your imagination ^_^

  • 2
    are these GPS tracks, by any chance? – Stev_k Nov 11 '11 at 10:39
  • 3
    It might help if you explain your use case. What do you want to do with all this data? – MerseyViking Nov 11 '11 at 12:36
  • 1
    And what software are available to you? Also I doubt the feature-extraction tag makes sense here. – blah238 Nov 11 '11 at 14:22
  • 1
    turns, angle, stops, time stopped, number of stops, walking?, cost comparison (other routes), time of day, traffic parameters, lanes available, construction, et al. – Brad Nesom Nov 11 '11 at 15:15
  • 1
    @radek, it sounds like what the OP really means by "features" is "attributes". – blah238 Nov 11 '11 at 15:37

Some (slightly) theoretical pointers:

  • Instead of focusing on attributes, one approach to the problem might focus on exploring characteristics of movement patterns. Those could be explored by calculating aggregated characteristics of movement or dividing your data into logical 'chunks' (for instance, daily trajectories of certain objects). At next stage you could look at the relations between all (or subgroups) of those trajectories. Have a look at some work done by Somayeh Dodge, especially Patterns of Movement wiki:

enter image description here

  • Patrick Laube has also done some interesting work on classification and mining of moving objects data that might give you some ideas.
  • Also, have a look at David Mountain PhD thesis: Exploring mobile trajectories: An investigation of individual spatial behaviour and geographic filters for information retrieval for some good pointers.
  • There is quite a lot of research emerging in the field of 'moving objects'. Google is your friend here if you want to discover more information about databases considerations and solutions, clustering, extracting trajectories, and many more. In case of your task, where retrieval seems to be focus - you might want to explore some efforts in trying to build specific database models that allow storage, retrieval and mining of spatio-temporal datasets.
  • Probably the most challenging (but also very interesting) aspect of the research is the prediction of movement. Have a look at paper in Nature Understanding individual human mobility patterns to have some inspiration. Again - Google will point you to many more projects in this field.
  • I think to work in this way, gave me your opinion: INDEXING: Having a lot of gps trace, I can compute a single trace (i.e. a map). After this, I'll analyze gps traces searching for interesting pattern (e.g. meet to discover parking, congestion to discover trafficated road), once finded, I can enrich the map adding this info. RETRIEVAL: Then, if I need to search for a trace that have covered a congestionable road, it will be finded quickly through map info, than I just to retrieve the trace that covered that road. What do u think about? It's a bad idea for IR aspect? – BAD_SEED Nov 16 '11 at 11:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.