Spatio-temporal trajectory models for target tracking

This paper presents generalized models for characterizing spatio-temporal target trajectories that have anomalous patterns. Stochastic context-free grammars (SCFGs) are the modeling framework used to represent anomalous events like circling behaviors and destination-specific trajectories. We propose a hierarchical tracking architecture to ensure legacy compatibility with existing trackers. The behavior of targets on the slower time-scale is captured through both positional features as well as movement patterns. Numerical simulations show a significant performance increase in probability of detection over competing hidden Markov model methods.