MULTI-AGENT ACTIVITY RECOGNITION WITH OBSERVATION DECOMPOSED HIDDEN MARKOV MODEL

摘要

The ability to recognize human activities using computer is important for any automatic surveillance system. Much effort has been done in the activity recognition of single person, such as gesture recognition. However, in many surveillance tasks, the scene usually involves several persons and their interaction plays an important role in understanding the meaning of the scene. This paper presents a new approach to model and recognize multi-agent activities from image sequences based on Hidden Markov Models (HMMs). In order to solve the problem of mapping agents from the dataset to the model, a new parameter which represents the role of each agent is added to the traditional HMMs.