An incremental learning mechanism for human activity recognition
Stavros Ntalampiras, Manuel Roveri · 2016
This paper proposes an incremental mechanism for the automatic recognition of physical activities performed by humans. The specific research field has become quite relevant as it may offer important information to areas such as ambient intelligence, pervasive computing, and assistive technologies. The works in the related literature so far assume the a-priori availability of the dictionary of activities to be recognised. This work is focused on relaxing that assumption by learning and recognizing the human activities in an incremental manner based on the acquired datastreams. To this end, we designed a learning mechanism based on hidden Markov models for recognising human activities among those of a dictionary. The major novelty of the proposed mechanism is its ability to detect the occurrence of new activities and update the dictionary accordingly. We conducted experiments on a publicly available dataset of six human activities, i.e. walking, walking upstairs, walking downstairs, sitting, standing, and laying, where the efficiency of the proposed algorithm is demonstrated.