2014 年 18 巻 3 号 p. 135-141
This paper describes an image-based agent's pose estimation method for supporting experiments on cricket-micro mobile robot interactions. Crickets modify their behavior selection tendency as a result of experience and the context based on the interaction with other crickets. Such a behavior selection mechanism with a tiny brain is a typical example to understand the adaptability of animals. Behavior observations in such interaction experiment sometime impose a burden on the experimenter. Therefore, in our previous works, we reported the development of a visual tracking method for two agents. However, it only tracked the positions of the agents. In this paper, utilizing our method, we also develop an image-processing algorithm for simultaneous pose estimation based on visual features related to the head part of each agent. We conducted estimation experiments based on video recordings of cricket-robot interactions and confirmed that our proposed method could estimate the pose of both agents. This method will help to efficiently analyze behavioral data during interaction experiments.