Maritime group motion analysis: Representation, learning, recognition, and deviation detection

This paper introduces new concepts and methods in the analysis of group motions over extended periods of time, and applies it to an example from the maritime domain. Group motion analysis includes these challenges: (1) represent complex motion patterns of multiple entities as they execute group maneuvers and behavior patterns; (2) learn these group motion patterns in a qualitative fashion, invariant to the number of entities participating; (3) recognize such learned maneuvers and behaviors as they unfold in future group tracking data; and (4) detect in real-time deviations of an entity from a recognized group maneuver or behavior, as such motion anomalies may indicate a vessel emerging from a group. The approach adopts a linear field theory to represent the relative motions of objects, in combination with learning & recognition methods that self-organize three evolving field parameters into 3D clusters that represent maneuver categories, and categories sequences that represent behaviors.