Analyzing the impact of different action primitives in designing high-level human activity recognition systems

Abstract

Designing human activity recognition systems, an integral part of any ambient assisted living environment, is an active area of research in the ubiquitous computing, wearable sensing, and computer vision communities. Yet most of the systems ignore human body motion and arm motion action primitives to recognize high-level human activities and are limited to object usage action primitives. Consequently, there is little understanding of the significance of these action primitives on the performance of activity recognition systems. In this paper, we comparatively assess the role of the object usage action primitives, body motion action primitives, and arms motion action primitives to recognize human activities of daily living. Our experiments show that the body motion action primitives and arms motion action primitives are vital to recognize the human activities that do not involve much interaction with the objects and the environment.