Effect of Activity Space on Detection of Human Activities by Domestic Service Robots

Activity recognition is a highly demanding aspect of human-robot interaction. Modern approaches mainly focus on the details of human body parts during activities but activity space is another differently utilized feature during each human activity. Therefore patterns in using activity space can be utilized as an additional feature to recognize a particular activity. This paper presents a vision based method to analyze how activity space is utilized during a selected set of tasks often implemented in domestic environments. For that, the space on either side of a human are divided into small regions for the ease of analysis. The size of these regions are scaled according to the dimensions of the particular person. Near front and side spaces around the human are divided into 50 small regions called 'zones'. Zones utilized by major body joints are analyzed under two approaches. The zones used by human during each activity is analyzed during the first approach. During the second approach, zones utilized are recorded with the frequency of visits to each zone throughout the period of observation. The system uses skeletal and relative spatial information to achieve this objective. Finally, these two approaches are compared with the help of experiment results. The capability of utilizing the patterns in these 'activity zones for task recognition is discussed. Applicability of activity space depending on the task is analyzed.