水果立體機械視覺分級系統之研究

Abstract

摘要 本論文發展出一個以平面影像估算立體深度的裝置與演算方法。本裝置由一組環狀鏡、圓錐鏡與照相機所組成。立體物體的360°影像經由環狀鏡反射至圓錐鏡，再投射到照相機的影像平面。利用一張這種影像與另一張正射影影像，可以估算出立體表面的空間座標，以作為立體影像重建。本研究以規則物體如長方柱、階梯圓柱、圓錐與番荔枝果實作實驗，水平方向的距離誤差為 2.5mm，垂直方向的距離誤差為 8 mm。以番荔枝果實而言，水平方向的相對誤差為 3% ，垂直方向則為 10% 。應用此種裝置與演算方法可以重建水果立體影像以便發展出依據形狀與大小的選別機械。

Abstract In this thesis a new stereo vision device and an algorithm to estimate depth from the stereo were proposed. The device could obtain omni-directional images of three-dimensional objects, it consists a pair of conic mirrors and a CCD camera. The conic mirrors are configured such that the 360° image of an object is collected and then projected onto the camera. Using this image and with a normal project image the depth of object could be estimated. Experiments were carried out on some regular shape objects i.e. cube, cone and staged cylinder and a fruit sweet sop. The estimated spatial distance errors were 2.5mm, and 8 mm for points on the same horizontal plan and for points at different vertical distance respectively that meant 3% and 10% relative errors for the width and depth of a sweet sop respectively. The device developed could be used for the fruit grading system based on size and shapes.