Was this page helpful?

Nearest neighbour tracker model

From $1

    Table of contents
    No headers
    This model is probably too simple for any real application, but it is useful for developing and testing the tracker infrastructure and GUI.  It works as follows:

    For each scan:

    • compute the distance from every new blip to the last blip in every active track
    • for any (track, blip) pair that would imply too high a speed, too rapid a turning rate, or too much
      relative change in blip area or intensity (per second), mark the distance as infinite
    • use this distance matrix to compute a minimum weight matching of tracks to blips; i.e.
      match blips to tracks in such a way that no blip is matched to more than one track, no track is
      matched with more than one blip, and the sum of all distances between tracks and their matched blips
      is minimized. (this uses Donald Knuth's implementation of the Hungarian algorithm for the
      assignment problem, coded in C)
    • every (track, blip) matched pair whose distance is infinite is unmatched, and the blip from that
      pair begins a new track

    That's it.  The model doesn't provide additional criteria for completing tracks, so a track completes when its last blip was a long time ago (see above).  Implicit in the model is that in the first scan with any blips, all of those blips begin their own tracks.

    Was this page helpful?
    Tags: (Edit tags)
    • No tags
    Comments (0)

    Powered by MindTouch Core