利用環場感測與二維影像分析技術作建築物走廊中自動車之航行與停車

Abstract

本論文中提出了一套利用電腦視覺與二維影像分析技術作建築物走廊中自動車之航行與停車的整合性方法。測試的平台是一輛配備有環場攝影機的自動車。為了應用影像角度與真實角度的不變性，我們提出了一套達到相機水平的校正程序。這種角度不變性使得車輛航行中之角度計算變得更容易。關於真實距離與影像距離的相對關係,我們提出了一個曲線映對的方法，利用所得的趨近函數便可達到距離的換算。對於航行路徑的學習，我們提出了一個利用特定路徑特徵的學習方法。而對於車輛導航方面，我們也提出了三個針對建築物走廊中之三種不同的區域類型的導航策略。此外，為了使車輛能平行停靠至走廊邊，我們也提出了兩種平行停車的方法，包過了正向與倒退停車。其中，我們運用了行駛曲線來重現停車軌跡以達到自動停車的目的。最後，良好的實驗結果也證實了本論文所提出方法與系統之可行性。

An integrated approach to autonomous vehicle navigation and parking in building corridors using computer vision and 2D image analysis techniques is proposed. A commercial vehicle mounted with a panoramic camera is designed as the test-bed. A calibration procedure for maintenance of the camera’s horizontal level is first proposed to achieve an invariant property of image angles with respect to real world angles. This property facilitates the computation of image angles during vehicle navigation. A curve fitting method is also proposed to compute real world distances from image distances using an approximation function. For navigation path learning, a technique is proposed, which is based on the use of certain path features. And for vehicle navigation, three strategies are proposed for vehicles to navigate in three different kinds of sections in building corridors. Furthermore, two parallel parking strategies, including a forward parking strategy and a backward one, are proposed to park a vehicle on corridor sides. And a new technique of using driving patterns for parking path guidance is also proposed to enable automatic parking. Experimental results show the feasibility of the proposed approach.