Abstract
To use the spatial memory system through human activities it is necessary to observe human behaviors in addition to the information on the usage of space. In this paper, we propose a method of motion recognition based on the time-series data of human posture. Similarity of motions is calculated by using the cosine similarity. This method can recognize motion regardless of individual difference of physical description. Humans are observed without any markers. Template data for motion recognition is generated based on the observation result.
