Human activity recognition in indoor environments by means of fusing information extracted from intensity of WiFi signal and accelerations

In this work, we propose an activity recognition system based on the use of a topology-based WiFi localization system combined with accelerometers for body posture recognition. The WiFi localization system is developed using a fuzzy rule-based classifier while the recognition of the body posture and its integration with the WiFi localization system is developed using two fuzzy finite state machines. These tools for modeling dynamical processes allow us to handle imprecise and uncertain data in the form of linguistic labels and fuzzy rules producing a linguistic description of the human activity. A practical application that consists of recognizing different activities of an office worker in her/his environment is developed. It yields high accuracy (83.7%, in average regarding all experimental trials). Interpretability and robustness of the proposal are also analyzed and alternative classifiers for the WiFi localization system are tested and compared obtaining competitive performance in terms of interpretability-accuracy trade-off.