SC²S Colloquium - November 3, 2017
|Date:||November 3, 2017|
|Time:||3:00 pm, s.t.|
Markus Fabry: Hierarchical Clustering of Trajectories with Sparse Grid Regression
The thesis introduces a clustering algorithm for trajectories aproximated with sparse grids. The learner builder extension of the SG++ toolkit is used to perform regression to obtain a functional representation of the trajectory data. The regression uses the hierarchical basis with adaptive refinement and modified basis functions to minimize the amount of grid points. Comparison and transformation of the trajectories into feature space can be accomplished with a variety of metrics, two of which are discussed in detail. To discover local groupings of trajectories, a hierarchical approach is used. The hierarchy splits the trajectories into more and more segments with each successive level. Each segment is then clustered separately using DBSCAN and compared to the corresponding segment of the previous level. If they are on average sufficiently different to the previous level, the splitting continues. Otherwise, the previous level is selected as the final split.