以環場視覺技術作室內停車場找車之擴增實境導引

Abstract

許多人都有一個相同經驗，那就是當我們要離開停車場時卻找不到自己的車輛，而在停車場裡迷路。為了解決這個找車的問題，我們提出了一個利用擴增實境技術幫助使用者快速尋找到自己車輛的導引系統。這個系統包含了自動記憶停好車輛的位置，以及用擴增實境方式導引使用者找到車輛的兩大功能，藉由多台安裝在停車場天花板上的魚眼鏡頭相機所攝得的影像，我們提出的系統可以做到四個主要的功能：使用者偵測、使用者追蹤、路徑規劃，以及擴增實境式的找車導引。 首先為了偵測進入停車場的使用者，我們提出一個動態建立背景的方法，把魚眼影像中的前景區域切割出來；也利用動態影像二值化的方法，從切割出來的前景區域中找到使用者的區塊。如此一來，使用者的位置就可以計算出來，並轉換到俯視的停車場地圖上。 為了追蹤使用者，我們提出了一多台攝影機換手(hand off)的方法，該法隨時切換監視使用者的攝影機，成為離使用者位置最近的一台攝影機。此外，我們利用卡爾曼濾波器(Kalman filter)和一些電腦視覺技術，讓所產生的人物行走軌跡更平滑，並顯示在停車場地圖上，藉此幫助使用者直接觀看手機上的地圖就可以得知自己處於停車場的位置。 為了導引使用者找到自己的車輛，我們提出了一個基於戴克斯特拉演算法(Dijsktra Algorithm)的路徑規劃方法；利用該法，我們可以規劃出一條從使用者現在位置到使用者車輛的最短路徑。此外，為了用直覺的方式導引使用者，我們提出了一個直接顯示導引箭頭在使用者手持行動裝置螢幕上的擴增實境方法，而且利用行動裝置裡電子羅盤的數值與下一個導引路線中目標節點的方向，提出了一個動態改變顯示箭頭的方法。 最後，本研究提出良好的實驗結果，證明本研究所提系統確實可行。

Many people have the experience of getting lost in a large parking lot when trying to find their parked cars before leaving the parking lot. To solve such a parked-car finding problem, an augmented reality (AR)-based guidance system is proposed to help a user to find his/her parked car quickly. The system has the functions of both remembering the location of parked cars automatically and guiding the user to find his/her car in an AR way. With multiple fisheye cameras affixed to the ceiling of the parking lot for image acquisition, the proposed system includes four major functions: user detection, user tracking, path planning, and AR-based guidance for parked-car search. Firstly, in order to detect a user entering the parking lot, a dynamic background construction method is proposed to segment the foreground regions in a fisheye image and a dynamic thresholding method is proposed to refine the user region in the segmented regions. Accordingly, the user’s location can be computed and transformed into a spot in the top-view parking-lot map. In order to track the user continuously in a large parking lot, a multiple-camera handoff method is proposed, which transfers monitoring responsibility among the cameras according to a criterion of minimizing the distance between the detected user’s location and the captured image center. Also used are Kalman filtering and some computer vision techniques for generating a smooth human track shown on the top-view parking-lot map, which helps the user to locate himself/herself in the parking lot by visual inspection. In order to guide the user to find his/her parked car, a method based on the Dijsktra algorithm for planning a guidance path is proposed, by which the system can plan a suitable minimum path from the current user’s position to his/her parked car. Also, in order to guide the user in an intuitive way, an AR-based guidance method which renders an AR-based guidance arrow for display on the screen of the user-held mobile device is proposed. Furthermore, a technique for updating dynamically the guidance arrow by utilizing the reading values of the electronic compass built in the mobile device as well as the location of the next node in the guidance path is proposed. Good experimental results showing the feasibility of the proposed system and methods are also included.