지능 공간 인터페이스를 위한 영상 기반의 다수 인간 행동 추적 방법

다국어 초록 (Multilingual Abstract)

This paper describes a real-time human tracking and human action recognition method that is for the interface with smart space. This method enables us to track multiple persons and recognize actions of each person by analyzing image sequences acquired from a fixed camera. In order to recognize all actions of several persons using video images, we need to track multiple persons and manage a list of the regions which is regarded to human appearance. Firstly, when given an image, regions must be segmented by using the MOG(Mixture of Gaussian algorithm) with a hierarchical data structure. Then persons are tracked by estimating the state of each region and assigning people to each region according to the regions state. The state of a region is defined as 5 states such as appearing, disappearing, continuation, merging, splitting. Finally we each MHI(Motion History Image) for the tracked people is constructed and actions are recognized by using a MLP(multi-layer perceptron). The proposed method is useful not only to track multiple persons in real time but also to manage human region data for action recognition and action tracking