Kurzfassung:
Labeling spatial trajectories, such as map matching, activity recognition, can ease the utilization of the imprecise and semantic poor spatial trajectories for location-aware applica-tions. This thesis studies the problem from a unified perspective using map matching and taxi status inference. Comprehensive probabilistic models are learned from the training data using a chain structure graphical model with feature selection, which are tested to be effective and feasible on a real world dataset.