基於快速光流法之單眼視覺自動避障與模糊控制停車導引智慧型輪椅機器人

Abstract

在本篇論文中，我們利用裝設在輪椅機器人上前後兩支攝影機，來導引輪椅 機器人於動態環境中自動避障與停車導引功能。在前方導引功能中，我們透過快 速光流法估測的方法讓系統可以短時間內偵測出環境障礙物的位置與目前的移 動狀態，並且系統可以依照障礙物離輪椅機器人距離的遠近動態規劃不同的區域 性避障路徑，來引導輪椅機器人自動閃避環境中的障礙物。 在後方停車導引的功能中，我們利用裝置在停車格附近的超音波感測器來感 測目前輪椅機器人的位置，當發現輪椅到達車格附近時，系統可透過無線WiFi 訊號，來啟動輪椅後方攝影機進行停車格偵測，並且我們仿照一般汽車的倒車入 庫方式，透過車格中線導引的二次曲線路徑，來引導輪椅機器人順利的停入車格 中。 本論文最後以實驗結果成功驗證出所提出的系統，在障礙物偵測上影像處理 速度達20 張/秒，準確率方面可達到93%，於環境中可以順利閃避障礙物，並且 順利引導停入車格。

The dynamic guidance for automatic obstacle avoiding and parking using the front and the rear camera on the wheelchair robot is investigated in this thesis. For the frontal guiding, the system can precisely detect the obstacle position and moving situation by the fast optical flow estimation. According to the distance between the wheelchair robot and the obstacle, the dynamic avoiding obstacle path could be established instantly. For the parking guiding, the wheelchair robot position will be distinguished from the surroundings sonar sensors transmitting the parking signals relying on WIFI and the wheelchair operates the rear camera to sense the parking space. In the end the robot is following the guidepath schemed by the least square curve fitting to instruct the wheelchair to implement the automatic parking function. In this thesis, the fast optical flow estimation has been proposed and addressed in a systematic manner in literature.Ultimately the approach is verified the efficiency (20fps) and accuracy (93%) by the automatic obstacle avoiding and parking functions in the dynamic environment.