Ontologies of Action and Object in Home Environment towards Injury Prevention
Satoshi Nishimura, Shusaku Egami, Takanori Ugai, Mikiko Oono, Koji Kitamura, Ken Fukuda · 2021
It is one of the critical applications for human-centric artificial intelligence that surveys the risky situation and inferring ways to prevent it by storing the situation information from surveillance cameras. Recognition of human activities in daily situations is an emerging topic in the computer vision domain. Significantly, the context information, such as objects involved in activities and the relationships between the objects and the activities, are attractive to improve the accuracy of the activity recognition task. However, the existing labels for actions and objects are not well considered for describing daily activities. This short research paper provides the ontologies of actions and objects in the home environment, so-called Primitive Action ontology, and Home Object ontology. The Primitive Action ontology contains a minimal set of primitive actions designed to abstract actions and discards objects and methods. The Home Object ontology has object types and properties such as affordance and attributes to describe daily situations. The properties represent both normal and abnormal effects, including intentional function and incidents in the home environment. We also discuss the prospect of using these ontologies as the conclusion of this paper.