人體肢節三維動作捕捉之電腦視覺技術

Abstract

本論文目的在於改進人體動作捕捉(Motion Capture)需要以光點輔助的限制，希望經由影像處理的技術來完成人體動作的捕捉。

本論文使用二台以上攝影機分別以不同角度拍攝，並做好攝影機的校正，使影像上的二維坐標點位，經計算成三維世界座標中的點位。透過影像處理，求得各肢節的方向向量，利用事先測量好的肢節長度重定肢節點位置，再利用逆向動力學(Inverse Kinematics)將重定肢節後偏差的肢節端點及角位修正回來，完成人體動作捕捉。

文中說明如何建立逆向動力學(Inverse Kinematics)的數學模式，及如何加入限制來讓人體動作捕捉能更加精確。最後用實驗驗證所提方法，分別以二張合成人體肢節數據及二組同步序列的拍攝真實人體影像做測試，說明本文方法的標準度

This thesis addresses the vision-base human motion capture without the need of placing the markers over the human joints.

Tow or more TV cameras are required to shoot the human motion from different viewpoints. Through the two or more views computer vision techniques are applied to find the 3D positions and orientations of the end_effectors including the head, two hands, and two foot tips, Together with the initial 3D orientation of limbs. Assuming the lengths of human limbs are known, a 3D human stick figure cam be constructed using the forward kinematics (FK). Due to the impression in the 3D limb orientations, there is a need of the application of the inverse kinematics (IK) to the constructed stick figure in order to bring the constructed stick figure to the correct configuration .

In the thesis, the mathematical formulation of the inverse kinematics is given . The modification of the Cyclic Coordination Decent (CCD) method is presented. Finally synthesis data and real image sequences are used to demonstrate the major features of the method.