Feature extractions for estimating human behaviors via a binocular vision head

We try to construct an estimation system of human behaviors, confining only the human behaviors that a robot vision head has met on, by extracting valuable information on the human behaviors from the binocular camera. Thus, in the present approach the human actions and objects related to the human behaviors are estimated by the images obtained from the camera, and their information are utilized for the human behavior estimation. Here, assume that, for the present estimation system of human behaviors, the information on a two-dimensional map that the human acts and behaves is known in advance and available.