本研究目的為設計機械手臂之雙眼視覺系統(binocular vision system)。此視覺系統主要功能為可以偵測、計算出目標物件於三維空間中的位置，並透過其所求出的資訊協助機械手臂順利到達所要求之位置，完成抓取物件的目標。本雙眼視覺系統包含裝置於機械手臂上之肩膀位置的兩顆CCD，以及一台筆記型個人電腦。本研究除了使用兩顆CCD外，並無加裝任何距離感測器(range sensor)進行目標物件深度量測。使用雙CCD擷取目標物件於環境中的影像，再利用各種影像處理(image processing)方法擷取目標物件之特徵點(feature points)，接著利用這些特徵點計算出目標物件中心點於影像平面上之像素距離，建立雙眼視覺成像模型；為了求出目標物件於真實世界中的正確位置，本研究使用座標轉換(coordinate transformation)公式推導出三維重建座標，將物件影像中的二維像素座標轉換成空間中之三維座標，以求出目標物件所在位置。換句話說，三維座標即是求出目標物件與機械手臂原點於空間中的三軸距離。所以，透過雙眼視覺系統所求得目標物件於空間中之三維座標資訊，可以使機械手臂完成以下三大任務：任務一、穩穩的抓取雞蛋並放入碗中；任務二、抓取裝有水的保溫瓶並將水倒入馬克杯中；任務三、精確按壓電梯面板上的按鈕。最後，經由實際的實驗與測試可證明本雙眼視覺系統是非常有效的。 The purpose of this study is to design a binocular vision system for a robot arm. The vision system can detect the target object and calculate its position in 3D space such that the robot arm can reach the target object according to the information from the vision system. The binocular vision system includes two CCDs which are installed in the shoulder of the robot arm and a person computer. Except two CCDs, there is not any other sensor. The CDDs capture the target image and then the image goes through some image processing to catch the feature points of the target object. Based on the feature points, the central point of the target object is obtained. A binocular vision model is constructed to calculate the 3D coordinate of the object from 2D image coordinate by the coordinate transformation formulas. In other words, the distances in 3D space between the target object and the original point of the robot arm is determined. Therefore, according to the calculated 3D coordinate of the target object, the robot arm can achieve three tasks. The first is that holding an egg and put it into a bowl; the second is grasping a bottle filled with water and pour the water into a cup; and the last one is pressing the exact number of elevator buttons. Finally, there are practical experiments to demonstrate the effectiveness of the binocular vision system.
