基於單眼視覺之擊球機器人設計與實現

Abstract

本論文基於一個二軸平面的機械手臂實驗平台，再添加單眼系統使系統成為具有機器視覺的機器人。利用此單眼系統透過影像處理過程，可執行對高速運動下的特定顏色目標物進行追蹤，其中使用的影像處理方式主要為改良式背景相減法與HSV的顏色辨識方法，再經過一連串濾波、形態學、邊緣偵測等方式將目標物擷取出來，並且利用切換影像處理區域，來提高整體執行速度。透過校正攝影機使影像平面能與真實空間整合，並且利用轉換公式與拋物線軌跡方程式來求得目標物飛行時軌跡參數。克服硬體上速度的不足，計算出最佳擊球點位置，使機械手臂可以在旋轉最小角度的情況下便可以擊打目標物，並利用即時持續更新影像擷取點來不斷地修正目標物飛行軌跡，再透過RS232傳輸線將資料傳送給高效能數位訊號處理器來即時規劃機械手臂路徑，使機械手臂能在移動情況下更新路徑且不斷地修正位置來擊打目標物。

This thesis is dedicated to the integration of a monocular vision system into a two - joint robotic manipulator, such that the robot can track a target with specific color in high-speed motion. Firstly, candidates of the target are identified from the images, by a combination of an improved background subtraction method and the HSV color recognition method. Then a sequence of image processing such as filtering, morphological processing, and edge detection are applied to correctly locate the target in the images To speed up these processes, the region of interest of the image is shrunk dynamically. Next a parabolic curve and pre-calibrated camera parameters are used to fit the trajectory of the target in the 3D space. Since the angular velocities of the robotic joints are limited, the best batting position is determined in the sense of least angular displacements of the robotic joints and is continuously updated whenever a new image is available. Through RS232 communication, the batting position is sent from the image process unit (a personal computer) to the motion controller (a high-performance DSP processor) of the robot for path planning such that the robot can hit the target at the desired position.