Human Activity Monitoring System Based on RGB-Depth Sensor

Abstract

This paper is related to the domain of human activity recognition in both depth images and skeleton joints. In this paper, for the detection task, a RGB-D sensor (Microsoft Kinect) is used. To obtain discriminative features for action detection, combination of a depth shape features from the 3D space and joints features are investigated. The detection and classification of such features is accomplished by the posture analysis technique, based on K-means and finally, activity recognition are performed by means of HMMs built on the set of known postures to improve performance and accuracy. The proposed system can be evaluated on a new dataset which contains five activities (standing, walking, sit down, lying and bending) and another public dataset MSRDailyActivity3D. The proposed system can be applied to the specific domain of healthcare system including home and hospital to keep older adults functioning at higher levels and living independently.