復雜幾何對象高精度數字化重建理論與方法

楊必勝　張帆　方莉娜　臧玉府　王晏民　侯妙樂　趙仁亮

摘 要：該研究的目標：研究多源、異構數據特征感知與融合的技術與方法，針對文化遺產的數字化復原需求，研究多源數據融合的文化遺產高精度重構方法，建立文化遺產的多尺度高精度表達模型，為文化遺產數字化存檔、虛擬修復、個性化展示等提供科學依據。為此，我們提出了一種基于空間曲率特征和形狀特征的高精度點云自動配準方法。在曲率特征點的計算基礎上利用點云聚類傳播和脊線提取技術，提取出信息量豐富的三維脊線，融合形變能量模型和最小二乘匹配方法，全面地描述空間曲線的形狀特征，同時利用曲線間的空間關系和最佳轉換參數篩選策略，完成高精度點云自動配準。通同時提出了一種基于興趣邊緣優化的壁畫影像與激光掃描數據非剛性配準和高精度紋理映射方法。利用激光掃描數據的強度信息生成強度影像，以壁畫彩色影像的興趣邊緣和強度影像的梯度場作為配準基元，在影像剛性配準基礎上，對每條興趣邊緣進行優化配準，然后以優化后邊緣的特征點為控制，構造影像之間的非剛性變換模型，完成壁畫影像與激光掃描數據的配準。針對缺失的幾何部件，該課題主要研究基于拉普拉斯算子的石質文物點云模型骨架線提取方法、不同尺度空間曲線相似度的分析方法來進行文物虛擬修復，提高交互式文物虛擬修復的可靠性。通過對敦煌莫高窟159號洞窟佛像的激光掃描數據（共22.4G，點云精度1 mm）和紋理數據的多次自動配準實驗表明該課題在文化遺產高精度數字重建的效果顯著。

關鍵詞：點云配準 三維重建 紋理映射 虛擬修復

Abstract：Digital heritage and archaeology have attracted a lot of attention of researchers in the world. Capturing digital representation of heritage objects can be used for the preservation， restoration， access， and scholarly study. The pipeline of 3D digitization includes 3D data capturing， 3D registration， geometric modeling and texture mapping. For capturing 3D data， we can use different methods to obtain the lidar data or imagery data from several views. The multi-views and multi-sources need to be registered and merged. Therefore， based on the curvature feature points， the three-dimensional crestlines with rich information are extracted by using the point cloud clustering method and ridge line extractiontechnology. The shapes of space curves are accurately described by merging deformation energy modeland least square matching method. In order to complete the high precision point cloud automatic registration， the spatial relationship between the space curves and the optimal selection strategy oftransformation parameters are explored. For realistic rendering， texture mapping is needed to map high resolution photos onto the geometric models. Texture reconstruction of 3D sculpture is explored by using non-rigid transformation. A method of cultural relic virtual restoration is proposed based on 3D fine model. Experiments using the laser scanning data and image data of the Buddha of DunhuangMogao grottoes 159 cave （22.4 G totally， the accuracy of point clouds with 1 mm）. Multiple automatic registration and texture mapping experiments show that our methods are able to reconstruct the cultural heritage with high precision.

Key Words：Point clouds registration；3D reconstruction；Virtual restoration；Texture mapping

閱讀全文鏈接（需實名注冊）：http：//www.nstrs.cn/xiangxiBG.aspx？id=48181&flag=1