Unsupervised discovery of basic human actions from activity recording datasets

Human Behavior Understanding (HBU) is a major challenge facing intelligent agents. Most approaches to solve this problem assume a recognition/detection context in which the agent/robot tries to match the perceived behavior to one or more predefined motion patterns (e.g. walking, running etc). A more challenging problem is discovering these motion patterns without apriori assumption about the motions in the data, their duration or their numbers. This paper proposes the utilization of a novel motif discovery algorithm based on the exact MK algorithm to discover basic actions in activity records. The proposed system was evaluated on real records of full body motions and is shown in this paper to achieve high accuracy compared with a recently proposed motif discovery algorithm applied to the same dataset.