Geometric Features-preserving 3D Surface Scanning And Reconstruction

3D surfaces present the real geometric shapes of 3D objects.Scanning and reconstructing 3D surfaces is one of the primitive topics in computer graphics and computer vision fields.Efficiently scanning a 3D target and faithfully reconstructing its surface benefit the theoretical investigations on many aspects including the analysis and editing of target shape,the detection and recognition of geometric features,geometry optimization and design and so forth.Meanwhile,a number of practical applications,such as the industrial measure,rapid prototyping,quality checking for products,medical visualization,movies and animations,virtual and augment reality,expect to obtain more faithful and abundant 3D surface models.For the past few years,3D printing is promoting the new industrial mode of personalized customization.It requires the convenient and effective capture of the target models even more urgent.Unlike capturing images and videos,3D scanning process is complicated and with high cost.Although handheld scanning using commodity depth cameras provides a new method for capturing the target 3D model,it still confronts several bottlenecks including the noisy depth images,a large number of redundant data,camera tracking losing.The reconstructed surfaces contain feature drift,detail blurring and surface deficiency,which limit the development of 3D geometry processing.To address the problems stated above,the feature-preserving 3D scanning and reconstruction are investigated systematically in this paper.We first explore the surface reconstruction via fusing sparse-sequence of depth images for handheld scanning using commodity depth camera,and then study the geometry completion for surface deficiency caused by target damage or occluded scanning views,finally seek the effective methods for enhancing and editing the geometric features for those reconstructed(both static and time-varying)models.Specifically,the main contents and the contributions of our work can be concisely stated as follows:(1)Propose the surface reconstruction method via fusing sparse-sequence of depth images for handheld scanning using commodity depth cameras.Based on the online analysis of camera trajectory it constructs the supporting subset for depth image sequence,which eliminates many redundant depth images and excludes the interference of those jittering frames.Our method introduces a refinement module to remove the heavy noise on the raw depth images and recover the geometric features.We finally complete surface reconstruction for the scanned target by fusing those refined supporting depth images sequentially into the truncated signed distance field of the target.Our method provides a flexible,robust and low-cost way to capture the 3D models for the target objects.(2)For the point cloud models with large deficiency we devise a shape-controllable geometry completion algorithm.It takes the hole-filling as an iterative propagation process of the hole-boundary.Each contraction of the hole-boundary is decomposed into two steps of normal propagation and position sampling so that the normal dissimilarity constraint can be integrated into the process of boundary contraction to control the recovered shape.Our method transforms the sampling of hole-filling on 2D manifold to the linear sampling by introducing the boundary control curve.For those deficient surfaces with sharp features,our method could generate both sharp and smooth hole-filling results.(3)Present a features enhancement method based on the cross-combination for time-varying surfaces.It first decomposes the input time-varying surfaces to obtain the continuous base surface sequence by designing a spatial-temporal coherent bilateral filter.Then the coherent detail sequence in temporal-domain is obtained by decomposing each frame individually just using a spatial bilateral filter.Finally,we devise a "cross-combination" strategy by adding the amplified temporal-coherent detail layers to the continuous base surface sequence to enhance the geometric features for time-varying surfaces meanwhile alleviate the jittering artifacts caused by the random motion of the target objects.The main three aspects of our work are centered on the topic of high quality surface scanning and reconstruction,which involves several important steps of 3D processing pipeline.These three aspects are closely related and provide the technical support for scanning and reconstructing the 3D surfaces.We test our algorithms on diverse 3D targets including objects with different shape complexity/material difference/distinct size,the real natural objects,the artificial models,as well as the mesh surfaces and the point cloud models et al.A great deal of experimental results verifies the feasibility and validity of our algorithms.Our algorithms could be applied to many fields associated with digital geometry processing,for example,the shape analysis and editing,3D scanning and printing,augment reality,animation and game et al.It will bring abundant 3D models for computer graphics community.