圓錐曲線理論為基礎之3D反推是利用圓盤之重要特性【Ma,1993】:空間中一已知半徑之平面圓的3D旋轉與位移資訊可由平面上的2D橢圓投影求得。利用此特性我們可利用攝影機拍攝空間中圓盤,利用影像處理方法來獲取圓盤投影之橢圓參數,再以圓錐曲線理論來反推出圓盤之3D空間座標,以作為相關應用之基礎【張朝舜,1999】、【韋立明,2000】。 過去數年中,本實驗室一直從事圓錐曲線理論為基礎之機器視覺應用於3D反推,但過去系統所遭遇到最大的限制為無法於自然燈光與複雜背景條件下追蹤圓盤橢圓投影,本研究利用整合多重影像資訊(EPIC)與結合隨機抽樣方法來估計橢圓參數(RANSAC)之橢圓追蹤技術來克服環境與燈光的限制【Vincze, 2001】,並針對整合多重影像資訊演算法之缺失進行改善,以提高於複雜背景條件下橢圓追蹤之準確度。最後以圓錐曲線理論為基礎之機器視覺技術發展出一套低成本、無線與直覺化之虛擬場景攝影機操控技術,提供使用者自然與直覺的虛擬場景攝影機操控方式,開啟未來虛擬場景操控介面另一發展方向。
The idea of conics-enhanced 3D recovery is based on an important property of conics theory: the 3D rotation and translation information of a planar circle with know radius could be derived from its elliptic projection. Using this property, we could use a camera to capture the movement of a planar circle and derive its 3D movement information based on conic theory【張朝舜,1999】【韋立明,2000】. We have been working on conics-enhanced machine vision approach for 3D recovery for 2 year. Our previous prototype system could not trace the elliptic projections of the planar circle under natural lighting and complex background. This research employs the edge-projected integration of cues (EPIC) method【Vincze, 2001】for ellipses tracking under natural lighting and complex background. The EPIC method, however, has problem when the ellipse tracked has similar gray-scale with the background or when the background is very complicated. We improve the EPIC method by adding another cue that gives each pixel on the tracking line a weight based on how chose the pixel is to the true ellipse boundary point at previous image frame. We then apply the conics-enhanced machine vision technique to develop a low-cost camera control method for more intuitive and smooth environment navigation.