利用電腦視覺在室內走廊環境作障礙物偵測與距離估測

Abstract

為了能使自動導航車在室內環境中，有效地閃避障礙物與估測出物體的距離，所以在本篇論文中，我們將建立一套雙眼立體視覺系統，針對對應問題( stereo matching )，利用立體視覺系統中，左右影像之間的特性，找出實際環境中障礙物成像至影像平面的投影點關係，以得到障礙物的實際深度，也就是得到相機本身與障礙物之間的距離關係，作為自動導航車的行進方向距離依據。找出對應點的關係是本論文的重點之一，另外為了減少尋找對應點的處理時間，需要先分辨出障礙物與環境，所以我們先針對走廊環境影像做分析，依據其影像特質，建立一個環境顏色database，再利用環境與障礙物的差異將障礙物切割出來，之後只要針對被切割後得到的障礙物影像找尋對應點。一旦這些點決定後，得到對應點的的位移差，則相機與障礙物間的距離便可得知。實驗結果充分顯示所提出的方法可行性及實用性。

In this thesis, we will develop a binocular stereo vision system to be used as an automatic navigational system. The main purpose of the vision system is obstacle detection and distance estimation for the collision avoidance of the automatic navigational system. A binocular stereo vision system includes a pair of left and right cameras. To find the correspondence points of stereo images (stereo matching) is a main problem of the stereo vision system. The pixels in the left and right images have similar structure. We can use these characteristics to find the conjugate pair corresponding to projection points of the scene points in the left and right images and then calculate the depth of scene points by perspective projection and triangular geometry. Determining the corresponding points is the main purpose in this thesis. And in order to reduce the time for determining corresponding points of stereo images, we need to identify obstacles in the environment first. Thus we analyze color space of corridor images. Based on the characteristic feature of images, we set up a database of environment color distribution. The color-segmentation method is employed to separate the obstacles from environment. Then the corresponding points of obstacles images are determined and the disparity of corresponding points are measured. As a result, the distance between cameras and obstacle is determined. Experimental results are provided to illustrate validity and applicability of the developed algorithm.