應用影像切割技巧建立室內通道的導航系統

Abstract

自走車導航系統必須具備在未知環境下擷取有用資訊的能力。在這篇文章中我們提出一個不需預先對環境瞭解的視覺導航系統，這個系統不但能使自走車朝正確的方向行駛還具備辨識轉角及門的功能。我們運用一改善後的區域成長法並利用地板與背景顏色不同的特徵來擷取地板的部分，此種方法是藉由成長的狀況作為是否過度成長的依據，此種方法成功的解決了過度成長的問題。接著我們以邊緣強化法取出我們要的邊緣，再以邊緣的長度、位置及期間的顏色來做為門的判定，最後以模組比對法來辨識走廊轉角的位置。因此我們可以得到室內導航所需的資訊。

Autonomous land vehicle (ALV) navigation requires to extract meaningful information in unknown environments. In this thesis, we present a vision-based navigation system in indoor environment without prior information. Our approach enables ALV to move toward the direction of vanishing point and to achieve doors recognition and junction detection. Because there are different colors in floor and wall in indoor environments, we extract the region of floors by region-based color image segmentation technique. And we propose a converge-guarantee region growing method to improve the drawback of the conventional region-growing algorithm. This algorithm can solve the problem suffers from over-growing by detecting the number of growing pixels in each growing layer. Doors are recognized by means of positions, geometric relationships of vertical edges and color values between two vertical edges. Furthermore, junctions of corridor can be identified by template matching technique. Finally, we extract junctions, doors, and vanishing point for navigation in corridors.