Modeling Human Activity Semantics for Improved Recognition Performance

Abstract. Activity recognition performance is significantly dependent on the accuracy of the underlying activity model. Therefore, it is essential to examine and develop an activity model that can capture and represent the complex nature of human activities precisely. To address this issue, we introduce a new activity modeling technique, which utilizes simple yet often ignored activity semantics. Activity semantics are highly evidential knowledge that can identify an activity more accurately in ambiguous situations. We classify semantics into three types and apply them to generic activity framework, which is a refined hierarchical composition structure of the traditional activity theory. We compare the introduced activity model with the traditional model and the hierarchical models in terms of attainable recognition certainty. The comparison study shows superior performance of our semantic model using activities of daily living scenario