The Mixed Ornstein-Uhlenbeck Process and context exploitation in multi-target tracking

A second-order Ornstein-Uhlenbeck (OU) Process, or Mixed OU (MOU) process, provides a stable and stationary generalization to the well-established nearly-constant velocity (NCV) model - the workhorse kinematic model used in the target tracking community. The MOU process is useful in many settings including long time-horizon simulations, multiple-model filtering of evasive targets, and hypothesis aggregation for improved track extraction. This paper clarifies the discrete-time target state covariance at birth and its relationship to the underlying continuous-time model. Further, we suggest additional fruitful applications of the MOU process including long-horizon prediction and ground-constrained tracking. We extend the MOU model to these context-aware settings and provide some evidence of its potential.