Visual context based infant activity analysis

This paper focuses on infant activities in one ambient smart home environment, and presents a novel method to analyze the infant videos. For infant activities, it is difficult to extract effective features of the infant in the videos. Moreover, infant activities are very flexible, whose pattern is not very obvious. In order to deal with these problems, this paper presents a method by generating a visual context and using a voting model to recognize each kind of activities. The testing infant videos are recorded in a home environment, which is much more natural than the environment of a laboratory. These testing videos are analyzed via the proposed method. We will demonstrate encouraging results on the infant activity recognition by our method.