以主成份分析法建立靈活之擴增實境座標系統

Abstract

擴增實境（Augmented Reality, AR）是將虛擬物件與真實物件在電腦產生之影像場景下做結合之技術。其中無標記的追蹤方法是使用真實物件當追蹤目標，將虛擬物件擺放至電腦影像場景中。本研究的目標是建立一個以電腦視覺為基礎之無標記追蹤方法。 研究方法使用主成分分析法（Principal Component Analysis, PCA）與平方預測誤差（Squared Prediction Error, SPE），從真實物件的顏色特徵取得閥值，進行追蹤。該方法使用顏色特徵將虛擬物件放置於電腦影像場景中。實驗結果顯示，該方法可用於偵測顏色特徵，估計攝影機之內部參數與外部參數，並在即時電腦影像中建立擴增實境座標系統。 實驗結果中，將真實顏色物件放置於電腦影像場景中，使用四個或更多的真實顏色物件，在部分真實顏色物件被遮蔽時，還是可以進行攝影機的姿態計算，建立擴增實境座標系統。使用有限制條件之計算公式來測量追蹤之精確度，藉由真實顏色物件在影像中之位置座標，計算出真實顏色物件於真實空間中之位置座標。實驗結果顯示，在邊長為30公分之立方體空間中，該方法的誤差於為0.3公分左右。實驗結果顯示，該方法是靈活、快速且精確。

The goal of augmented reality (AR) research is to combine virtual objects with real objects in computer generated scenes. Markerless tracking methods use real object features to place virtual objects in computer generated scenes. The goal of this research study is to create a vision-based Markerless tracking method. The method uses principal component analysis (PCA), squared prediction error (SPE) calculation, and feature extraction thresholds to extract color features from real objects. The method uses color feature to place virtual objects in computer generated scenes. Experimental results show the method can be used to detect color features, estimate camera pose parameters, and create 3D coordinate systems in real-time video camera images. Experimental results were achieved by adding real colored objects to computer generated scenes. Four or more real colored objects were added to computer generated scenes, to handle occlusion effects for different camera pose parameters. A constraint function was used to measure tracking precision. The constraint function projects real object space and compares real object space feature points to image space feature point to determine tracking precision. Experimental results show that the method can be used to place virtual objects with, 0.3 cm precision in a 30 cm x 30 cm x 30 cm computer generated scene. Experimental results show that the method is flexible, fast, and precise.