自動車在建築物內之路徑學習、規畫與導航

Abstract

本論文中提出了一套利用電腦視覺技術在建築物內進行自動車導航的方 法。所提出的模式學習技巧可應用在建構環境模型、學習路徑迴轉方法以 及抽取環境特徵供模型節點確認之用等方面。在建構完成的模型上，我們 用戴克斯查(Dijkstra)演算法來規畫最短的路徑。本論文中所提出的自動 車導航方法，主要是基於將自動車保持在所抽取出之路面中央的觀念，同 時藉著判斷路面的狀況來達到避碰的目的。論文中歸納出一組有系統的避 碰決策規則。當偵測到障礙物時，可依循該組規則決定出避碰的方向。另 外可以透過查表的方式得知避碰所需轉角的大小。本論文中所提出的方法 ，已在一輛實際的自動車上實作，並成功地在建築物內之走廊與房間內進 行導航多次，證實了此方法的可行性。 An integrated method for autonomous land vehicle (ALV) navigation inside buildings using computer vision techniques is proposed in this study. Techniques of model learning are proposed to create the model of the environment, to learn turnaround processes, andto learn scene models for node verification against the model. With the graph model set up, path planning of the minimal cost path is achieved by Dijkstra's algorithm. The main idea of the proposed navigation approach is to keep the ALV on the generated central path of the extracted route and avoid possible collision with obstacles by judging the condition of the route. A set of rules for collision avoidance is induced systematically. If there is any obstacle detected, a turn angle is obtained by consulting look-up tables, and a moving direction is determined according tothe proposed rules. The proposed approach has been tested on a prototype ALV and many navigation sessions inside buildings have been performed successfully to confirm the feasibility.