Human Activity Recognition In Time Series In PIR Sensor Networks With Reinforcement Learning For Smart Lighting

Recognizing human activity from IoT events is essential for automatic light scene setting. A challenge with IoT systems is the diversity among customers. For addressing this diversity a reinforcement learning method is explored that enables for each customer the learning of a model for classifying context of time series. The method is a Bayesian method and it is of interest as a prior is put on the states that are to be predicted. Results on activity recognition will be given for a household and the relative feature-importance of time-of-the-day versus motion will be analyzed.