Three-Dimensional Image Information Media. Animation Synthesis of Virtual Fish by Motion Extraction from Video.

In this paper, estimation method of fish position and posture from video using object matching technique is proposed. In the object matching, 3D reference model is divided into three parts, head, body and fins to accelerate the object matching process. NURBS (Non-Uniform Rational B-Spline) is employed for body model in order to represent flexible fish body postures using a small number of control points. The control points of the model is obtained by nine motion parameters, i.e. the center of gravity, the inclination of a fish's head and the angles between skeletons which represent the posture of the body. Some basic behaviors are segmented from the obtained 3D positions and postures by using the characteristics of the motion of fish. Finally, a novel locomotion is generated using the segmented basic behaviors according to the segment connection rule.