An Unsupervised Approach to Recognize Activities of Daily Living by Interacting with Objects in the Home

摘要

In this thesis, a new machine learning technique for recognizing activities in the home with a set of simple state-change sensors is introduced. In this sensor environment, a set of simple state-change sensors is installed on the objects to detect residences' behavior during their day. The proposed method uses an ontology model to store the information about the sensors. The algorithm uses Fuzzy Logic to generate temporal knowledge based on the discovered patterns among the low-order temporal events. Unlike prior works, the new algorithm is easy to extend by introducing a new set of sensors or a new set of activities to recognize. The result has shown that it is possible to recognize the activities without having prior knowledge about the individual by using partial truth. The accuracy of method is between 20% to 80% depending on the activity and criteria of evaluation. The Activity recognition could be used for variety of applications such as preventive health-care, designing buildings, security systems, etc.