利用電腦視覺技術作室內環境學習、路徑產生及自動車導引

Abstract

本論文提出一套使自動車能在建築走廊中自動作環境學習、路徑產生及自動導引的整合技術。我們運用電腦視覺技術在走廊內作自動車定位。使用到的環境特徵包括走廊中的踢線及角點。我們提出一種模式比對的方法來找出最可信的環境特徵配對，並利用此一環境特徵配對來作自動車定位。我們也提出一套環境學習的方法來記錄自動車在學習階段中所經過的環境特徵，以期達成環境學習的目的。此外並提出一套路徑產生的方法，找出一條能讓自動車作平穩航行的合理路徑。最後，我們製作一輛實際的自動車作測試，並達到了自動車在走廊中作平順且安全航行的目標。經過了多次成功的實驗，我們證明了此套方法的可行性。

An integrated approach to automatic model learning and path generation for vision-based autonomous land vehicle (ALV) guidance in building corridors is proposed. Computer vision techniques are utilized to locate an ALV in corridors. Used environment features include baseline segments and corners on walls. The ALV location work is accomplished by a matching scheme for finding possible matching corner pairs or baseline segment pairs. Furthermore, strategies for model learning are proposed in order to acquire automatically the environment features that the ALV pass through in a learing stage. And techniques of path generation are also proposed to generate paths automatically for guiding the ALV in indoor environments. Finally, a real ALV was constructed as a testbed, and smooth and safe navigation sessions can be achieved. Lots of successful experiments confirm the feasibility of the proposed approach.