Building representative velocity profiles using FastDTW and spectral clustering

The use of map based representative velocity profiles allows to predict the future state of a vehicle. The suggested approach is based on Fast Dynamic Time Warping. Spectral clustering is used to distinguish velocity profiles. Applying abstraction can significantly reduce computation time with a minor effect on cluster allocation. Outlier removal increases the quality of cluster identification. The approach was applied to the road network of Munich, to prove the universal applicability.