Daily Human Activity Recognition in Smart Home based on Feature Selection, Neural Network and Load Signature of Appliances
Nadia Oukrich · HAL (Le Centre pour la Communication Scientifique Directe) · 2019
A smart home is a standard residence that was improved and equipped with all kinds of sensors and effectors in order to provide services to its residents. One of the most key subjects and input to several smart home applications (e.g. healthcare and home security) is the recognition of activities of a resident’s daily living. Being able to automate the activity recognition from human motion patterns is challenging because of the complexity of the human life inside home either by one or multiple residents. To surmount all databases complexity, several algorithms of features selection and machine learning were tested in order to increase the human activity recognition accuracy. Another major challenge to cope with isto reduce the costs of maintenance and installation of sensors at home. These sensors, despite their modest costs, are generally out of reach of most people. To overcome this challenge, we used another approach based on household appliances recognition as sensors that detect human interaction with appliances and resident movements. This study aims to solve the complexity of human activity recognition and increase accuracy by proposing two different approaches. The first approach is based on recognizing human activities using ambient sensors (motion and door sensors), neural and deep neural network combined with several feature selection methods in order to compare results and define the influence of each one in learning accuracy. The second approach is based on load signatures of appliances presented using an Intrusive Load Monitoring in order to identify the most accurate classifier suitable for appliances recognition. Once determined, the next phase is to know resident activities through appliances recognition. Each part of our methodologies is thoroughly tested, and the results are discussed.