透過您的圖書館登入
IP:3.144.154.208
  • 學位論文

基於三維建模與多視角影像擷取之物體三維定位及追蹤技術

Model-Based Object Localization and Tracking with Multi-View Image Captures

指導教授 : 林惠勇
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


在本論文中,我們提出利用物體虛擬三維模型,從多個視角拍攝物體,對影像 中的物體進行空間三維定位及追蹤技術。給定模型的初始的位置,將虛擬模型投影 在影像上,利用模型所投影的輪廓線去找出影像中與其匹配的物體特徵線以後,透 過非線性姿態計算技術,找出物體在三維空間中的位置資訊。當物體在移動過程中 有遮蔽的話,影像特徵擷取不佳將導致無法找到物體特徵線,或著匹配錯誤使得姿 態計算錯誤。為了解決這個問題,我們提出了使用多台相機同時從不同角度拍攝物 體。當有相機影像發生影像特徵擷取不佳的情形時,改以其他視角所計算的姿態經 由相機座標轉換後取代該相機的姿態,用來作為進行下一張影像的模型初始姿態。 以及提出一個在單台拍攝的情況下,當影像視角發生特徵擷取不良的情況時,能透 過不使用影像特徵資訊的姿態預測法,建造一個新姿態作為姿態結果。在我們的實 驗中,我們進行距離精準度及旋轉精準度實驗來驗證在真實空間中的定位準確性, 且實驗在單台拍攝與多台拍攝之下,物體在遇到遮蔽以後的追蹤情形,來驗證多視 角的姿態座標轉換方法及單視角的新姿態預測法的可行性。

並列摘要


In this thesis, we present a model-based object localization and tracking technique from multi-view image captures. We project the 3-D model to the image, use the nonlinear pose computation technique to compare the model line and the object feature line to find the object location in the real world. When the object is moving, it may cause the object with partial occlusion and let the tracking result wrong. To solve this problem, we use multi- view images captured from various viewing angles to track the object. Using the camera coordinate transformation between cameras, the good pose will transform to replace the bad pose. In single camera the object with partial occlusion, we use our approach to the previously obtained precise poses to create a new pose and replace the bad one. In our experiment, we include an examination of the positioning accuracy of the proposed method, and the object tracking in single and multi-camera with serious partial occlusion in our camera coordinate transformation method the pose prediction is satisfactory.

參考文獻


erless tracking for augmented reality: the virtual visual servoing framework,'' IEEE
Transactions on Visualization and Computer Graphics, vol. 12, no. 4, pp. 615--628,
[3] B. D. Lucas, T. Kanade, et al., ``An iterative image registration technique with an
[5] M. Vincze, ``Robust tracking of ellipses at frame rate,'' Pattern Recognition, vol. 34,
[6] M. Isard and A. Blake, ``Contour tracking by stochastic propagation of conditional

延伸閱讀