本論文提出了一套機械手臂之應用系統，本系統設計一具有5個自由度(degree of freedom)之機械手臂，機械手臂前端有夾具可以抓取空間中的目標物，並以雙攝影機實現目標物追蹤。為了由影像追蹤目標物，作者提出一個以立體視覺為基礎之機械手臂定位方法，首先透過攝影機校正方法(camera calibration)求出雙攝影機之內外部參數予以校正，並結合雙眼立體視覺(binocular stereo vision)求出物體在世界座標系統中實際重心位置。雖然物體的重心位置經過校正，但在估算與攝影機中心的距離時仍有些微誤差，因此本系統利用多層感知機(Multilayer perceptrons)修正此誤差。接著計算出順向運動學(Forward kinematics)所得之端點座標與目標物實際位置之誤差值，並以其誤差值當作基因演算法(Genetic Algorithms)的適應函數 (fitness function)推算出機械手臂中各馬達的角度，達到機械手臂定位之目的。結果驗證部分，我們將目標物放在隨機放置在機械手臂的工作範圍內，並透過上述演算法將目標物夾取到目的地。This thesis presents a vision-based robot arm with 5 degrees of freedom and an end-effector attached to the end of the robot arm for grasping an object. In order to locate a target from the image, a stereo vision-based robot arm system is implemented. A stereo calibration algorithm is adopted for estimating the two cameras’ intrinsic and extrinsic parameters. With the estimated parameters, the stereo-vision system can estimate the 3D position of the object in the world coordinate system. A trained multi-layer perception is then used for compensating the location estimation errors incurred by the inaccurate parameters estimated from the calibration procedure. In the following, the Genetic Algorithms (GA) is adopted to solve the forward kinematics problem in order to compute the angles of each motor. The performance of the robot arm was tested by several real-life experiments of griping a target at different positions.
