Abstract:Feature Matching has long been the basis and a central topic in the field of computer vision and image processing. SIFT (scale-invariant feature transformation, by David G. Lowe), due to its advantages of invariance to image scale and rotation, and robustness to a substantial range of affine distortion and change in viewpoint, has been attracting the attention of many domestic and foreign researchers over a decade. Rapidity and accuracy are very crucial for stereo pairwise image matching in applications such as 3D reconstruction. First, in order to accelerate the speed and promote the accuracy of matching, this paper proposes a novel method based on SIFT called approximately consistent in orientation (ACIO), which depicts the spatial location relationship of two matched vectors between stereo pairwise images (SPI), and therefore improves the accuracy of matching efficiently by avoiding the wrong correspondences. Secondly, this paper analyzes the structure of standard K-d tree (SKD-tree) and proposes a new one with hierarchical structure, named HKD-tree, which partitions the feature sets of SPI into stripes in terms of ACIO constraint and builds map between them. By reducing the search space, the matching speed increases greatly. Thirdly this work presents an efficient and fast matching algorithm based on ACIO and HKD-tree. Extensive trials based on a benchmark data set show that the proposed approach outperforms the state-of-the-art methods in matching speed with slight promotion in accuracy. Particulary, it is one order of magnitude faster than SKD-tree and also several times against the recent CasHash method.

Abstract:Regularized Boolean operations have been widely used in 3D modeling systems. In recent years, Boolean algorithms based on triangular polyhedron show the distinct advantages aligning with the development of graphic hardware, as their outputs can be processed by graphic hardware directly. But most existing methods rely on localized evaluation strategy over constructive solid geometry (CSG) tree perform regularized set operations. As a result, these methods cannot guarantee robustness while synchronously keeping high efficiency, because a facet may repeatedly split up in the splitting phase and the facets classification is carried out between the split polyhedrons by triangulation. In this paper, a novel algorithm is presented to realize robust, exact and fast regularized Boolean operations through global evaluation of CSG tree. The algorithm is comprised of two steps:adaptive lazy splitting and globalized facets classification. The two steps aim to optimize splitting and facets classification phases of the regularized Boolean algorithms on triangulated polyhedrons respectively. In the adaptive lazy splitting phase, a lazy splitting strategy is applied to the whole CSG tree by coping with all intersection cases of triangular facets in order to eliminate the accumulation of number errors. In the meantime, an adaptive octree is employed to speed up the intersection test process. In the globalized facets classification phase, to ensure the accuracy of classification, the classification method is always executed between the split facet and the original input polyhedrons by divide and conquer algorithm. The performance of classification is further optimized by combining the grouping classification strategy and the octree. Experimental results demonstrate that the proposed approach cannot only guarantee the robustness of Boolean computations but also achieve better performance than existing approaches. Thus, the algorithm offers wide-ranging usage in for interactive modeling systems, such as digital sculpture, and CAD/CAM.

Abstract:Freedom curve/surface reconstruction with normal constraints is crucial in optical reflecting surface design. In this paper a binary code based genetic algorithm for knot optimization scheme is proposed to reconstruct a B-spline curve that not only approximates the data points but also meets the corresponding normal constraints. First, the constrained optimization problem is transformed into an unconstrained optimization problem by means of penalty function method. Then, the binary code based genetic algorithm (GA) is applied to find the best knot vector after establishing a suitable fitness function. Finally, adaptive generation of optimal knot vector and iterative evolution result in a satisfactory reconstructed curve. Since knot vector is non decreasing,and genetic algorithm may disrupt the order of knot vector in searching for the optimal knot vector, a process is also built to adjust variables into disordered bounded variables in the fitness function. Test results and a comparison with the traditional least square method as well as modern particle swarm optimization method show that the proposed scheme for reconstructing B-spline curve with normal constraints is superior and effective on arbitrary shape of discrete data set.

Abstract:Efficient modeling of minimal surfaces is a challenging problem and hot topic in the field of geometric design and computation. Taking boundary closed polylines, this paper proposes a general framework to construct discrete minimal surfaces with quadrilateral meshes. First, the mathematical definition of discrete minimal surface with quadrilateral mesh is given from the intrinsic differential-geometry metric of surfaces. Next, based on the length-preserving boundary projection method, quad-mesh generation approach and non-linear numerical optimization technique, a novel framework is presented to construct discrete minimal surfaces with quadrilateral meshes from a described boundary closed discrete polylines. Finally, the effectiveness of the proposed approach is illustrated by several modeling examples. The results show that the proposed method can achieve high-quality modeling of discrete minimal surfaces and provide potential usage in architecture geometry.

Abstract:This article presents a fast 3D color full-body capture system, named Avatar 1, which can scan a subject in 1.5 seconds and automatically provide the full-body model within 1 minute. It uses eight self-made structure-of-light scanners which are built based on structured lighting technique and have a measuring accuracy of 1mm. The eight scanners are configured to effectively cover the full-body surface of an adult, four scanners are arranged to capture the head, while four target the body. After calibrating the scanners in the offline step, it can capture the real subject and process all partial scanned data in the same coordinate. Through the global registration and seamless texture stitching steps, it robustly reconstructs a high-quality full-body textured model, which has 1mm geometric accuracy and photo-realistic color information. The experiments demonstrate that Avatar 1 achieves state-of-the-art scanning and processing speed under high-quality result requirement.

Abstract:Fast, low-cost, and accurate three-dimensional scanning technology is a hot research topic in computer vision. First, this paper introduces the technical parameters and measurement principle of the Kinect v2 (Kinect for windows v2 sensor) and analyzes the advantages and disadvantages of Kinect v2 comparing to the first generation Kinect (Kinect v1). Secondly, considering the characteristics of Kinect v2 depth data noise, it designes a denoising method for depth data, and applies this method to the 3D reconstruction system. Finally, it implementes a real-time accurate 3D reconstruction system based on Kinect v2. Experimental results show that the proposed 3D reconstruction system can acquire an accurate 3D object model in real-time.

Abstract:Natural phenomena simulation is an important research area in computer graphics domain. How to obtain realistic natural phenomena by simulation in an efficient way is the main purpose of this research field. The physics-based modeling method is a traditional way to solve this kind of problem. Along with rapid development of acquisition devices, the image-based method attracts a lot of research attention and interest. This paper chooses liquid as research object. Main research achievements on the topic of image-based reconstruction are summarized. Focusing on dynamic water surface, a method of dynamic water surface data acquisition and reconstruction is proposed. First, a dynamic water surface data acquisition system is designed and built in order to obtain image sequences of different water surface phenomenon on multi-viewpoint. Next, feature points on each captured images in sequences are extracted. Then after feature point matching process, the mapping relationship between feature points and 3D points are built. Finally, dynamic water surface reconstruction results are obtained by using the height field and normal of 3D point combined with the optical refractive property of water. Experimental results show that this method can generate 3D reconstruction results rapidly. Satisfactory visual effect is also achieved facilitating this application to be used in computer games, visualization in scientific computing, and virtual simulation in medical area.

Abstract:Rapid architecture modeling often has many conflicting goals, including high precision, high efficiency, simple user interaction, and low demand for user domain knowledge. To solve this problem, a component-based multi-level parameter modeling method is proposed. First, probabilistic network is used to describe the high-level parameters of the architecture and components as a friendly user interface. Then, rule library is used to describe the low-level parameters of the components and to guide the generation of low-level parameters from high-level parameters. Finally, construction template is used to describe the construction procedures of the architecture. Based on these groundworks, an automatic Huizhou architecture building system is implemented to provide rapid Huizhou architecture modeling with real construction process and high precision. The experimental results demonstrate the practicality and effectiveness of the proposed method, showing that the system delivers simple and friendly interaction, good scalability, and good user experience.

Abstract:In the literature of computer graphics and computer vision, single-image based 3D architecture modeling is a hot research topic. Focusing on repetition detection of urban architecture images, a novel algorithm for reconstructing detailed 3D architectures is presented for interactively generating large number of repetitive structures in an effective and convenient manner. The new approach consists of three steps. First, according to the parallel or vertical relations of sketch lines interactively drawn by the user, the coarse architectural model is reconstructed using an energy function optimization. Upon the texture information of the single input image, the horizontal and vertical repetitions on each architectural face are then detected and classified based on their bounding boxes. Next, according to the user inputted sketch lines of typical detailed structure, the 3D information of detailed concave or convex structures is calculated based on the projective relations between detailed structures and basic body of architecture. Finally, the other same type of repetitive structures are reconstructed automatically and their 3D detailed architectures can be generated. Experimental results show that the proposed method can conveniently and effectively reconstruct 3D detailed architectures from single image with high visual quality.

Abstract:This paper presents a data-driven method for rapid 3D garment modeling, which can efficiently generate new garments by examples. First, a semantic segmentation is applied to simultaneously segment and label the components in given 3D garments using shape analysis of garment styles. Second, the garment components are clustered into a garment component library with four categories including tops, bottoms, sleeves, and accessories. Third, a continuous style description of the 3D garments, characterized by the ratio of area and boundary length, is constructed to recommend components that can be regarded as a new 3D garment. The final new suitable garments are produced by optimizing two component meshes. Experimental results show that the presented method is able to satisfy the new requirements of online 3D model collections while ensuring the efficiency of 3D garment modeling.

Abstract:Traditional cloth and garment simulation is normally based on the mesh model where cloth and garment are generally rendered according to the lighting, texture and color setting, which cannot fully reflect their photorealistic appearance. This paper proposes a novel fiber-based cloth and garment simulation algorithm that facilitates photorealistic appearance. The algorithm first uses the sphere set constraining the particle perturbation to express the contour of fiber, and then simulates the fiber-based yarn reconstruction helped by the dynamics and kinematics principle of yarn twisting process. Next, it combines the organization structure of the yarn warp knitting with the mass spring model to construct the cloth model based on the fiber blocks. Finally, it builds up the hybrid bounding volume hierarchy of the human body model to speed up the collision detection between the body model and the garment composed of the fiber based cloth, which helps achieving satisfactory dynamic simulation. Experimental results show that the presented method is valid and effective, as it can reflect the detail in the fiber level of the cloth models and express the geometric texture and the photorealistic appearance of garments.

Abstract:The rapid modeling and controllable animation of large scale water surface and waves have always been hot issues in computer graphics research. However, due to the complexity and irregularity of natural waves, the existing simulation methods of waves cannot make a good tradeoff between computational efficiency and realism. In this paper, a technique is presented to extract features of the wave shape and generate controllable wave animation rapidly. Taking the captured video of outdoor costal wave as input, this technique first uses mathematical morphology algorithms to resolve wave's mask and skeleton from water video sequences as wave's features. Then it employs these features to control the shape of waves by reusing of height field to generate various wave animations. Consequently, the new method can produce controllable swell wave moving patterns under simple interaction with a small computational cost. Experiments show that the presented method can generate natural wave deformation effect with a simple and intuitive control process.

Abstract:To address the problem of merging the multiple meshes of the same human ear, which is scanned from different perspectives, this paper proposes a novel approach-patch removing, for merging 3D ear meshes. First, kd-tree is used to segment the two registration meshes of 3D ear into un-overlapping area and overlapped area. Then, the un-overlapping area and overlapping are segmented into blocks according to their connectivity. Next, the redundant patch in the overlapping area is discarded, thus establishing the points on the boundary of blocks. Finally, to merge the 3D ear meshes, the mesh blocks in the un-overlapping area and the rest of the overlapping area are stitched along the boundary points. Experimental results show that, comparing with other works, the proposed algorithm can merge the 3D ear meshes more effectively with higher computational efficiency.

Abstract:Porous metal fiber sintered felt is a type of new functional materials. This paper focuses on implementing the digital design of its multi-scale morphology. First, considering the self-affine fractal characteristics of the microtopography of machined metal surfaces, a previously developed mathematical method combining Weierstrass-Mandelbrot fractal geometry and triply periodic minimal surface is extended. In addition, the marching cubes algorithm is optimized according to the structure characteristics of sintered felt, so as to improve the efficiency of establishing its geometrical model. Meanwhile, a parameter representation method is introduced to drive the fractal TPMS model to adjust and control the morphology of sintered felt. The sample analyses warrant the higher efficiency of the proposed method and the ability of modeling and controlling multi-scale morphology of sintered felt. The effectiveness of the proposed model is also validated through numerical simulation and comparisons with other methods. The proposed approach can be directly used to describe the multi-scale morphology of other functional materials, thus facilitating the development of the corresponding numeric simulation technology.

Abstract:Containment is an important safety device for the protection of nuclear reactors. Visualization for measurement data or simulation data of the containment can help discover potential security issues. This paper introduces a mesh generation method and a physical data visualization method by leveraging cylindrical coordinates to represent the shape of the containment. By setting one-dimensional attribute of the cylindrical-coordinate system, the cross-section and longitudinal-section representations of mesh cells are illustrated. By using an Octree structure to store cells, the presented approach achieves good space performance and time performance. By setting one or more dimensional attributes, line and section representations for the physical data are obtained. Experimental results show that the proposed method can depict the running of the containment in real-time, simulate the physical laws of the containment effectively, and therefore ensure the safety of nuclear reactor.

Abstract:This paper proposes a multi-layer surface reconstruction method based on a special oriented status octree. The method is designed to handle point sets with missing normal information and with holes. First, the octree cells distributed on holes are detected by active contours. By using hidden point removal (HPR) operator, the inside or outside status of each corner of octree cells are calculated, and the mono-oriented status octree is constructed. Then the normal direction of the initial points set inside the status octree is determined by in-cell normal detection method, and the parallel multi-layer surface reconstruction from k-nearest neighbors is carried out by using the status octree. The proposed method improves the construction efficiency while guaranteeing the construction quality.

Abstract:This paper presents a novel feature based mesh model representation to represent the missing high level information of the traditional 3D mesh model. The application value of this representation is demonstrated by employing it in the model manipulation approach. Its feature structure, which consists of feature line, feature patch and feature group, is designed to describe the shape, constraint and semantic information of the model. As a result, it constructs a hyper-mesh beyond the original mesh model. With the topology and constraint information between features, the model can maintain its shape and structure properties during the editing. Meanwhile, as the modifications are limited in the local area, the manipulation approach is more efficient.

Abstract:An optical image based multi-granularity follow-up environment perception algorithm is proposed to address the follow-up environment perception issue from indoor to outdoor in the field of rapid 3D modeling. The algorithm generates multi-granularity 3D point cloud models which perfectly fit the ground-truth according to different types of optical image. A probabilistic octree representation is proposed to uniformly express the 3D point cloud models. Finally, the expected TFPOM is generated through dynamic ground-truth fitting at any granularity, and probabilistic octree representation of multi-granularity point cloud models are dynamically fused through implementation of Kalman filter along with the camera trajectory. Benefiting from pruning and merging strategies, the proposed algorithm meets requirements of multi-granularity fusion and multi-granularity representation. As a result, the storage space of environment models can be effectively compressed and robust follow-up environment perception can be achieved, which are essential in environment model based visual navigation and augmented reality. Experiment results show that the algorithm can generate multi-granularity TFPOM which perfectly fits ground-truth in real time with fewer errors in model based navigation on platforms, such as wearable devices, that are equipped with multiple optical sensors and low computing capability.

Abstract:In this paper, a novel reconstruction method based on free-form deformation (FFD) and External Axes Projection (EAP) is presented to improve the surface smoothness effect of 3D reconstruction. The contours of the slices are implicitly embedded in a higher dimensional space of distance transforms. In this implicit embedding space, reconstruction is formulated as follows. First, an arrangement of the planar slices is computed to support the approach. The arrangement consists of cells, and each cell consists of two adjacent contours. Second, the branching problem is converted into one-to-one case by the external axes projection. Next, computing direction for each cell of the arrangement is decided by the length energy. Then, in each cell a B-spline based free-form deformation is used to establish the correspondence between the adjacent contours. Finally, the contours are stitched together based on the correspondence. The key advantage of such framework are:(1) it naturally deals with contours of arbitrary topology, and it preserves shape topology; (2) the established one-to-one correspondences between two adjacent contours can guarantee the surface is continuous and no intersection; and (3) this framework is highly parallel. Experimental results show that the proposed approach performs well and can handle complicated situations.