利用擴增實境與環場電腦視覺技術作園區行車導覽之研究

Abstract

本研究利用架設在車頂上的環場攝影機及投射於擋風玻璃上的影像，偵測特殊地標後將所得到的資訊直覺地顯示於擋風玻璃上，成為完整的擴增實境戶外園區導覽系統。駕駛者可以利用投射在擋風玻璃的擴增影像看到目前經過的左右兩邊的建築物及其名字。 為了讓每個地標都有導覽的資訊，首先要建立園區導覽地圖，在本研究中，我們提出了環場攝影機與PTZ攝影機的座標轉換方法，以此建立單一地標的區間地圖。在單一地標區間地圖建立完成後，我們也提出把區間地圖合成完整的園區導覽地圖的方法，獲得每個地標在導覽地圖中的位置關係。 在本研究中，此環場攝影機取像系統能夠大範圍地監看任何角度的影像畫面，並偵測環境中之地標資訊。為了要偵測此地標資訊，我們利用圓形地標在環場影像中會得到橢圓形狀的特性，並研究此橢圓面積與地標位置之關係。更詳言之，我們利用影像處理技術，偵測出地標在影像中的中心位置，並計算出地標的真實世界座標，得出監控車和地標之間的方向與距離。 同時，我們利用平板投射在擋風玻璃上的影像達成擴增實境的應用，發展出顯示目前車子所在的建築物資訊的方法。本研究也提出得到平板放置的位置與投射的影像位置的方法，讓車內的人了解成像的位置。最後我們利用圖表遍歷法及光流法，偵測出監控車轉彎的方向以及在導覽俯視圖中如何得知目前的位置。 實驗結果顯示所提出戶外園區導覽系統確實可行。

In this study, an augmented-reality based tour guidance system for use in park areas using a video surveillance vehicle and computer vision techniques has been proposed. When a user drives the vehicle in a park, he/she can get from the system tour guidance information about the names of the nearby buildings appearing along the way on the two roadsides. The building names are displayed on the iPad, which then are projected onto the windshield for the user to observe on the driver’s seat, in a sense of augmenting the scene seen through the car windshield by the projected building names. To implement a system of this augmented reality function, at first, a PTZ camera and a two-camera omni-imaging device equipped on the vehicle roof are used for acquiring PTZ-images and omni-images, respectively, for the purposes of guidance map construction and vehicle localization. Guidance map construction is carried out by use of the PTZ camera to measure feature points on nearby buildings. Vehicle localization during vehicle driving is accomplished through 3D image analysis using omni-images acquired of a series of red circular-shaped landmarks attached beforehand on along-way flat roadside objects (sidewalks, pedestrian chairs, etc.). To detect red circular-shaped landmarks around the video surveillance vehicle during park navigation, image analysis techniques, including YCbCr color modeling, region growing, approximation of a circle in an omni-image by an ellipse, elliptical shape fitting, etc., are applied to segment out the corresponding elliptical-shaped landmark in the omni-image. Also proposed is a method for obtaining the center point of the ellipse in the omni-image to compute the 3D data of the detected landmark for vehicle localization. Furthermore, methods for calculating the accurate position of the iPad image projected on the windshield, computing the accurate position of the vehicle in the guidance map, and showing the names of the nearby buildings are proposed. Finally, also proposed are methods for analyzing the vehicle turning direction by motion vectors produced by optical flow analysis, and for generating a guidance map graph for use in path planning for each tour in the park. Good experimental results are also shown, which prove the feasibility of the proposed methods for augmented reality based tour guidance in park areas.