Meta Communication Lab.

Director: Prof. Katsuhiko Shirai

We are doing close studies of the methods used to communicate with robots smoothly as well as for the recognition of human faces under various conditions. We analyzed and evaluated the methods used to control the robot in expressing non-verbal information. As the first step in recognizing the motion of a human face, we tried to extract the human face under various conditions, and to detect facial poses and eye-gazes.

Outputs of non-verbal information with a CG robot

Non-verbal Communication

We focused on the use of non-verbal information in human-robot dialogue to realize natural communication as human beings. First, the output timing of non-verbal information during interactive dialogues between human beings was analyzed. Moreover, we analyzed influences of output timing by controlling the dialogue with a CG robot. As a result, we clarified the differences of the output timing according to various types of non-verbal information. From experiments of communication between a human and the robot, it was confirmed that the use of non-verbal information with the appropriate output timing at the beginning and the end of utterances makes speaker-change smooth, which is the same as in human-human dialogues.

Results of Pose and Eye-Gaze Detection

Facial Pose and Eye-Gaze Detection

When a human instructs operations to a robot, or works cooperatively with a robot, it is necessary to inform the robot of the human's intentions and interests. The motion of a human's face and gaze is important non-verbal information that indicates a human's intention and interests. Then, the recognition of such motion can lead us to smoother communication with the robot. We are doing research on facial poses and eye-gaze detection.

Target image

Result of face extraction

Result of pose detection
Result under various backgrounds or light conditions

Motion Recognition of the Human Face

We are trying to integrate facial pose detection with face extraction, in order to recognize the motion of the human face. These figures show some of the results of face extraction and pose detection, and some results under various backgrounds or light conditions. We can extract the human face under various conditions, and detect facial poses as well.

Content Top    

Copyright by Humanoid Robotics Institute, Waseda University. All rights reserved.