ZEMCH 2015 - International Conference Proceedings | Page 245

not just presence, but also inferring the number of occupants inside a space through multi-sensor
readings. As in Mamidi’s work (Mamidi et al., 2012) where a combination of multiple sensors (i.e.
motion, temperature, humidity and CO2) readings was fed into different machine-learning approaches such as SVM reporting high standard results from the models evaluated.
One important research objective in the human pattern modelling in buildings is the development of activity pattern recognition (APR) models, which are mainly focused on the modelling of
occupant activities (ADL classification) in domestic scenarios. User activities are estimated from
sequences of sensor readings by identifying sensor event patterns to infer which activity occupants are conducting. The ultimate goal for activity models would be the creation of ambient
intelligence environments to adapt the systems in place to real user needs. These include diverse
applications such as digital houses or healthcare systems which try to monitor patients or elderly
people and discover any abnormal behaviour that could represent a symptom of illness or accident. It is normal therefore for these scenarios to incorporate a high sensor density and diversity
(the latter depending of the complexity of the activities intended to model). From the TV to media
devices, appliances, communications, security or the mentioned healthcare issues, all of them are
common applications for APR models. Existing Approaches of Activity Recognition Smart home
is a concept frequently related to APR, since the main goal of these models is to regulate and automate different equipment at home, seeking the ultimate user satisfaction. One of the pioneering works was proposed in (Si et