User Activity Recognition via Kinect in an Ambient Intelligence Scenario

A great number of sensors are nowadays available, this has stressed the need for new approaches to merge low-level measurements to realize what facts they refer to in the real environment. Ambient Intelligence (AmI) techniques exploit information about the environment state to adapt the environment itself to the users’ preferences. Even if traditional sensors allow a rough understanding of the users’ preferences, ad-hoc sensors are required to obtain a deeper comprehension of users’ habits and activities. In this paper we propose a framework to recognize users’ activities via the Microsoft Kinect. The approach proposed here takes advantage of the position of some human body parts estimated by using Kinect depth information. In our system, significant patterns of joints (i.e., postures) are discovered by applying a clustering technique and then classified by means of a multi-class SVM. Each activity is then modeled as a sequence of known postures by using HMMs. A prototype has been implemented by connecting the Kinect to a miniature PC with limited computational resources. Experimental tests have been performed on a dataset we collected at our laboratory and results look very promising.