HO2: A new feature for multi-agent event detection and recognition

摘要: In this paper, we present a new feature to model class of events that consist complex interactions among multiple entities captured by tracks and inter-object relationships over space time. Existing approaches represent these using features measure only pairwise between at time, such as relative distance speed. Due the limitations entity relationship descriptors, is mainly defined recognized rule-based approach. The feature, Histogram Oriented Occurrences (HO2), captures all interests in terms configurations HO2 encapsulate tracks, context environment into spatial distribution characterizes corresponding event. compact structured descriptor for capturing multi-object relationships. We demonstrate its value event detection recognition standard statistical clustering classification techniques.