Multi-level Resolution Features for Classification of Transportation Trajectories

We explore the use of filter-like multi-level resolution features of a positional trajectory for classification. Our approach is time and location agnostic which increases generality. Several filter types are discussed and used in feature extraction including moments and wavelets. Previous work by Bolbol et al. is extended to incorporate these features and results are shown for each framework and filter type. We attempt a 6-way classification of mode of transportation from GPS trajectories obtained from cell phone handsets. Our primary contribution is that our approach can classify an entire trajectory, regardless of its length, overcoming a deficiency in other approaches which require trajectories to be segmented into equal length parts. We achieve >60% accuracy split between 6 classes where the 'random' feature accuracy is <;28%, an 'informative' gain of over 30%..