Academic journal article International Journal of Electronic Commerce Studies

Activity Analysis with Hidden Markov Model for Ambient Assisted Living

Academic journal article International Journal of Electronic Commerce Studies

Activity Analysis with Hidden Markov Model for Ambient Assisted Living

Article excerpt

ABSTRACT

In an Ambient Assisted Living (AAL) project the activities of the user will be analyzed. The raw data is from a motion detector. Through data processing the huge amount of dynamic raw data was translated to state data. With hidden Markov model, forward algorithm to analyze these state data the daily activity model of the user was built. Thirdly by comparing the model with observed activity sequences, and finding out the similarities between them, defined the best adapt routine in the model. Furthermore an activity routine net was built and used to compare with the hidden Markov model.

Keyword: Activity Analysis, Ambient Assisted Living, Hidden Markov Model, Forward Algorithm

(ProQuest: ... denotes formulae omitted.)

1. INTRODUCTION

There are many papers about the hidden Markov model: The author in1 explained the basic definition of a hidden Markov model with applications. Paper 2 builds a daily activity model of the elderly with a hidden Markov model. The authors in paper 3 used Bayesian posterior probability to choose the states to merge and to stop merge. The authors in paper 4 introduce the theory of the hidden Markov model and illustrate the applications in speech recognition. In paper 5 a segmental k-means algorithm is used as objective function to estimate the joint likelihood between the observation data and the Markov state sequences. A new type of hidden Markov models is presented by the author in paper 6. In the new models the current state is related both to the preceding state and the preceding observation. The authors in paper7 search the dynamic transition probability parameters A(t) and propose a moveable hidden Markov model. The hidden Markov model is used to learn the behavior of the user for building an automated system which was introduced in paper 8 and 9.

1.1 AAL, Project ATTEND

Ambient Assisted Living (AAL) aims to improve the life quality of the elderly with the help from modern technology and prolong the period of independent living in their own home. But the elderly have their own problems because of aging, such as action obstacles, memory disorder..., how can a modern technology system adapt to the elderly? 10 A system will be developed at the project ATTEND (AdapTive scenario recogniTion for Emergency and Need Detection) that extends the period of independent living of the elderly. An intelligent, adaptive network of sensors will be installed in the living environment of the user, in order to thoroughly observe the activities and behavior of the user. Then the user's activities and behavior model will be learned by the system over a time period. Based on the learned model if unusual behaviors or activities of the user are observed the system will send an alarm signal to the caregiver. Neither camera nor microphone will be used in the whole system and especially in project ATTEND because of privacy issues of the elderly. Furthermore there will be no sensor to wear on the body or to be activated by the user. The only sensors used are a motion detector or door contactor.

1.2 Contribution

In this paper at first the huge amount of raw sensor data was translated to state data. Then using a hidden Markov model, a forward algorithm builds the activities model of the user. Thirdly observed activity sequences are matched with the best adapt routine in model to find the similarities between them. Finally an activity routine net was introduced in order to compare with the hidden Markov model.

2. TRANSLATE RAW DYNAMIC SENSOR DATA TO STATE

The raw data come from a motion detector installed in the living room of the user. If there is any "movement" of the user, the motion detector will send value "1" to the controller. Otherwise the sensor value is "0". Because the activities of the user are random so the raw data is randomly distributed over time intervals during the day and in a huge quantity (hundreds sensor data every day). In order to reduce the data amount and to get more typical activity states the raw data will be translated to state data in predefined time intervals 10. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.