ActRec - A Non-monotonic Reasoning Tool for Activity Recognition

• Main Author: Partonia, Saeed
• Format: Others
• Language: English
• Published: Umeå universitet, Institutionen för datavetenskap 2015
• Online Access: http://urn.kb.se/resolve?urn=urn:nbn:se:umu:diva-99647

Human activities usually have a motive and are driven by goal directed sequence of actions. Recognizing and supporting human activities is an important challenge for ambient assisted living. Human activity recognition has a wide scope of application areas, e.g., aged care support, health care, smart homes, natural disasters and energy efficient urban spaces. Different techniques have successfully been applied to infer human activity, including machine learning and data mining. These data driven techniques work well within a particular domain and situations in which they are initially set in. However two main drawbacks with such methods have been observed in the literature: they are domain dependent and also require large amount of data annotation for model training. Hence, different authors have argued for exploring complex activity recognition techniques that not only rely on data but also involve domain knowledge is necessary. Against to this background, in this project, we explore non-monotonic reasoning technics in order to capture domain knowledge in terms of action specification languages. By considering an action specification language, called CTAID, and Answer Set Programming, we propose and develop a system called ActRec system which takes background information into consideration and operates independently from the environmental factors. We also explore a novel definition of activity which is used in the implementation of ActRec.