– Recognizing daily life activities and human behaviour from contextual information is a challenging task. The purpose of the research work in this paper is to develop a system that can detect indoor and outdoor daily life activities of low entropy mobile people such as elderly people and patients with regular routines using non-intrusive sensor and contextual information.
– A framework is proposed that utilises a hierarchical approach in which high-level activities are divided into sub-activities and tasks and recognises the high-level outdoor and indoor activities of daily life. Tasks are recognised at lower level from sensor data and then used by the “activity recogniser” at higher level to recognise the high-level activities. For outdoor activities recognition, wireless proximity data are used, whereas for indoor activities, object usage data obtained through radio frequency identification sensors are used.
– For outdoor tasks, results have shown 100 per cent recognition for experiment 1 and a decrease in recognition from 100 to 82.7 per cent, respectively, for experiment 2-9 due to increase in the entropy of individual tasks. Outdoor activity recognition ranges from 84.1 to 100 per cent. For indoor tasks, generating alternative tasks sequences approach effectively recognised the single tasks that were conducted with objects without any order. Average indoor activity recognition rate remains above 90 per cent. The reason why this approach is able to detect the activities without their distinct features is the planning capability of the Asbru that is used in the modelling of high-level activities.
– The novelty of this research work is a framework that utilises different types of sensor data and recognises both indoor and outdoor daily life activities of individuals.
