A Study on Video Conference Through Virtual Reality Scene Based On Face Tracking Using 3D Sphere Motion Technique

• Main Authors: Jyh-Keng Wu, 吳志耿
• Other Authors: Rong-Chin Lo
• Format: Others
• Language: en_US
• Published: 2002
• Online Access: http://ndltd.ncl.edu.tw/handle/38147685213461526712

碩士 === 國立臺北科技大學 === 資訊工程系碩士班 === 90 === In the paper, several techniques have been proposed to head detecting and tracking in real time. Using color saturation and hue of the skin, the face can be segmented from environment. Following, we divide eye ranges and mouth range from the face, and then compute the center points of these ranges as possible tracking positions. After this, we can decide the tracking feature points from characteristics of eyes and mouth. Via image preprocessing the edge of the segmented face are also found to construct 3D sphere model. Then, the head tracking can be performed by cross correlation. Using the size of face projected into image plane, the distance of face from screen can be calculated out, and using forward and inverse perspective transformations to calculate out face rotation angle. After this, we can use the proposed 3D sphere model to allocate the pose of head. Via network, we also build a virtual reality videoconference by the proposed head tracking techniques. Without any additional marks on the face, we can transmit the facial expression and detected 3D head location and orientation of the users via network and construct the 3D virtual reality videoconference. Several demonstrations show feasibility and practicality of all the proposed approaches.