Research On Hierarchical Hidden Markov Model-based Activity Recognition

With development of science technology, and progressive improvement of living standards, people hope the living environment more intelligent, which provides the development of smart home more space and makes research on smart home become hotspot in academic research. The key of intelligent activity monitoring system is to make the system has the ability of learning and understanding user's behaviors. So learning and recognition of activities is the most important step towards smart home.During the process of learning target user's behaviors, relationship characteristics between users and surrounding environment are numerous and complex, user's behaviors are usually unstable as well. In order to more efficiently resolve these problems in activity recognition, we need to select appropriate behavior characteristics to describe user's behaviors. Based on expression of activities, how to choose a reasonable and efficient activity recognition model becomes a key factor to realize activity recognition.This paper defines user trajectory position changes using Rcc-D as the user's behavioral characteristics. Training and testing data sets are obtained by generalizing the data of user trajectory position changes. This article presents a method to model complex human behavior hierarchically in complex environment. We propose a use of Hierarchical Hidden Markov Model (HHMM) that has recently been extended to handle shared structures, for representing and recognizing a set of complex indoor activities.The main results of this paper can be summarized as follows:1. After analysis relative activity recognitions, the author gives studies on resent situation and existing problems which shows position and importance of activity recognition in smart home.2. Propose a new position relational model Rcc-D and an algorithm of automatically generating Rcc-D composition table. Use Rcc-D to represent use's position changes as characteristics of activities in3D environment space.3. Use Rcc-D to represent user's position changes as characteristics of activities and present a method based on hierarchical hidden Markov model to recognize activities. 4. With experimental simulation, this paper compares hierarchical hidden Markov model and hidden Markov model to show the proposed method can improve the accuracy of activity recognition.