Human Activity Monitoring System Based on RGB-Depth Sensor

This paper is related to the domain of humanactivity recognition in both depth images andskeleton joints. In this paper, for the detection task, aRGB-D sensor (Microsoft Kinect) is used. To obtaindiscriminative features for action detection,combination of a depth shape features from the 3Dspace and joints features are investigated. Thedetection and classification of such features isaccomplished by the posture analysis technique,based on K-means and finally, activity recognitionare performed by means of HMMs built on the set ofknown postures to improve performance andaccuracy. The proposed system can be evaluated on anew dataset which contains five activities (standing,walking, sit down, lying and bending) and anotherpublic dataset MSRDailyActivity3D. The proposedsystem can be applied to the specific domain ofhealthcare system including home and hospital tokeep older adults functioning at higher levels andliving independently