Human activity recognition based on the activity categorization using depth data for "biofied building"

The number of single households is gradually increasing in Japan. The majority of them are consisting of elderly people. The "Biofied Building" that can monitor and control living spaces will help those people in many aspects. One of the key items to be acquired is the information about the residents activity. In this paper, we suggest an activity recognition method for two categories of activities, "Single type" and "Multiple type." We use the consecutive depth data obtained by Kinect. The R transformation, variance, PCA, LDA, and k-means classification are used for classification. Ten activities data sets consisting of six "Single type" and four "Multiple type" conducted by eleven subjects are used for verification.