Geometric Methods for Shape Recovery from Line Drawings of Polyhedra

Using structural geometric arguments, Whiteley showed that a line drawing is a correct projection of a spherical polyhedron if and only if it has a cross-section compatible with it. We here enlarge the class of drawings to which this test applies, including those of polyhedral disks, possibly with perforations. This extension is helpful, as it makes the test applicable to verify and reconstruct drawings from usual scenes with opaque objects.The presented results rely on geometric constructions, thus offering an alternativeapproach to line drawing interpretation, complementary to the algebraic-combinatorial treatment given in the classic work by Sugihara.Lluís Ros was born inVilafranca del Penedès (Catalonia), on June 30, 1968. He is currently a research scientist at the Institut de Robótica i Informàtica Industrial of the Spanish High Council for Scientific Research, where he holds a Ramón y Cajal position since February 2003. He received a M.Sc. degree in mechanical engineering in 1992, and a Ph.D. degree (with honors) in advanced robotics in 2000, both from the Polytechnic University of Catalonia. His research interests are in geometry, with applications to machine vision, robot kinematics and computer graphics.Federico Thomas was born in Barcelona, on October 5, 1961. He received the B.S. degree in telecommunications engineering in 1984, and the Ph.D. degree (with honors) in computer science in 1988, both from the Polytechnic University of Catalonia. Since March 1990, he has been a research scientist in the Institut de Robòtica i Informàtica Industrial of the Spanish High Council for Scientific Research. His research interests are in geometry and kinematics, with applications to robotics, computer graphics and machine vision.     

Shape-From-Silhouette Across Time Part I: Theory and Algorithms

Shape-From-Silhouette (SFS) is a shape reconstruction method which constructs a 3D shape estimate of an object using silhouette images of the object. The output of a SFS algorithm is known as the Visual Hull (VH). Traditionally SFS is either performed on static objects, or separately at each time instant in the case of videos of moving objects. In this paper we develop a theory of performing SFS across time: estimating the shape of a dynamic object (with unknown motion) by combining all of the silhouette images of the object over time. We first introduce a one dimensional element called a Bounding Edge to represent the Visual Hull. We then show that aligning two Visual Hulls using just their silhouettes is in general ambiguous and derive the geometric constraints (in terms of Bounding Edges) that govern the alignment. To break the alignment ambiguity, we combine stereo information with silhouette information and derive a Temporal SFS algorithm which consists of two steps: (1) estimate the motion of the objects over time (Visual Hull Alignment) and (2) combine the silhouette information using the estimated motion (Visual Hull Refinement). The algorithm is first developed for rigid objects and then extended to articulated objects. In the Part II of this paper we apply our temporal SFS algorithm to two human-related applications: (1) the acquisition of detailed human kinematic models and (2) marker-less motion tracking.     

特征点维度的静态场景三维重建运动目标剔除

基于语义分割的图像掩膜方法常用来解决静态场景三维重建任务中运动物体的干扰问题,然而利用掩膜成功剔除运动物体的同时会产生少量无效特征点.针对此问题,提出一种在特征点维度的运动目标剔除方法,利用卷积神经网络获取运动目标信息,并构建特征点过滤模块,使用运动目标信息过滤更新特征点列表,实现运动目标的完全剔除.通过采用地面图像和航拍图像两种数据集以及DeepLabV3、YOLOv4两种图像处理算法对所提方法进行验证,结果表明特征点维度的三维重建运动目标剔除方法可以完全剔除运动目标,不产生额外的无效特征点,且相较于图像掩膜方法平均缩短13.36%的点云生成时间,减小9.93%的重投影误差.     

基于反投影注意力网络的遥感影像超分辨率重建

近年来, 在诸如环境监测等一系列工作中, 遥感影像得到了广泛应用. 然而, 目前卫星传感器观测到的影像往往分辨率较低, 很难满足深入研究的需要. 超分辨率(SR)目的是提高图像分辨率, 同时提供更精细的空间细节, 完美地弥补了卫星图像的弱点. 因此, 本文提出了一种反投影注意力网络(back-projection attention network, BPAN)用于遥感图像的超分辨率重建, 该网络由反投影网络和初始残差注意块两部分组成. 在反投影网络中, 通过迭代误差反馈机制计算上下投影误差指导图像重建; 在初始残差注意块中, 引入初始模块融合局部多级特征为重建详细的纹理提供更丰富的信息, 以注意模块自适应地学习不同空间区域的重要性, 促进高频信息的恢复. 为了评价该方法的有效性, 在AID数据集上进行了大量的实验, 结果表明, 本文的网络模型提升了传统深度网络的重建性能, 且在视觉效果和客观指标方面有明显提升.     

基于子空间差异的投影聚类算法

针对传统K-means型软子空间聚类技术中子空间差异度量定义的困难问题,提出一种基于概率距离的子空间差异表示模型,以此为基础提出一种自适应的投影聚类算法。该方法首先基于子空间聚类理论提出一个描述各簇类所关联的软子空间之间的相异度公式;其次,将其与软子空间聚类相结合,定义了聚类目标优化函数,并根据局部搜索策略给出了聚类算法过程。在合成和实际数据集上进行了一系列实验,结果表明该算法引入子空间比较可以为簇类学习更优的软子空间;与现有主流子空间聚类算法相比,所提算法大幅度提升了聚类精度,适用于高维数据聚类分析。     

空地融合的城市道路基础设施数字化技术研究

城市道路基础设施三维模型的重构,在城市道路BIM应用与数字化领域具有重大意义;针对城市交通基础设施数字化重构的需求,对车载激光扫描与无人机倾斜摄影采集技术进行综合运用,提出一套从信息采集、空地点云配准、点云分割到三维重构的完整技术方法;首先使用车载激光扫描技术和无人机倾斜摄影技术对交通基础设施信息进行采集,并使用运动恢复结构算法(SfM,structure from motion)生成基础设施空地点云;其次使用迭代最近点法(ICP,iterative closest point)对空地点云进行精配准,然后利用基于PointNet网络的方法对融合点云进行语义分割;最后对分割出的交通基础设施对象进行三维重构;提出的空地融合的城市交通基础设施数字化技术能够高效地实现交通基础设施重构,为城市交通基础设施数字化提供基础、为后续交通专业领域的应用研究提供便利.     

基于深度图像的三维场景重建系统

针对计算机图形学和视觉领域研究热点--三维场景重建,首先分析了 Kinect v2 (Kinect for Windows v2 sensor)获取深度图像的原理,说明深度图像噪声的来源。然后根据获取 深度图像的原理设计一种算法对点云采样范围进行裁剪。其次对点云离群点进行去除,填补点 云孔洞,以提高重建质量。常见的三维场景重建大都采用了 KinectFusion 的一个全局立方体方 案,但只能对小范围内的场景进行重建。对此设计了一种对大场景进行点云匹配的 ICP 算法。 最后对点云进行曲面重建,实现一套低成本、精确的针对大场景的三维重建系统。     

基于图像的重建点云离群点检测算法

基于图像重建出的三维点云模型通常会包含许多离群点,这些离群点可能孤立存在或密集聚集在一起形成点簇,也可能分布在模型周围甚至附着在模型表面。通过一种检测方法很难有效滤除多种分布状态的离群点,因此,提出了综合的离群点监测算法。首先通过空间距离剔除与模型主体较远的离群点,并通过构建空间拓扑关系加快离群点搜索速度;然后利用边界匹配法,将较小点簇分别与最大点簇进行对比,滤除模型周围离群点簇;最后采用改进的K-means算法,根据RGB颜色值特征对点云数据进行聚簇分类,结合已识别的离群点,检测和滤除附着在模型表面的离群点。仿真实验结果表明,此方法能够有效滤除点云模型中多种分布状态的离群点。     

基于深度卷积神经网络的图像重建算法

视频或者图像在传输过程中,可能出现随机性的误码、突发性的误码、传输中的丢包等等,对解码出的图像数据也会有严重的影响.本文提出了基于深度学习的图像重建算法：一种基于图像背景预测生成模糊区域内容的无监督图像重建神经网络模型.为了重建出逼真的图像,神经网络模型需要既理解整个图像的内容,又为缺失的部分重构出一个合理的假设.损失函数包含标准像素级重建损失和对抗损失,在训练卷积神经网络模型时,能够更好地处理图像中的结构细节产生更清晰的结果.通过实验可以发现本文设计的深度卷积神经网络模型与基于样本插值的算法相比在图像重建中有着较好的效果.     

最优方向耦合字典学习的遥感影像超分辨率重建

针对遥感影像超分辨率重建问题,提出了一种改进联合字典学习的超分辨率重建模型。利用最优方向字典更新算法进行耦合字典对的学习,将由低分辨率字典学习得到的稀疏系数传递至高分辨率字典学习空间,形成高、低分辨率字典对,重建得到高分辨率遥感影像。该算法通过优化,实现训练样本自动截取,通过验证实验表明：与已有的经典算法相比,提出的算法定量评价指标有明显改善,同时,在字典学习过程中所需时间远少于现有经典算法,大大提高了遥感影像重建的效率,其重建影像更加清晰,几何纹理结构更加明显,证明了该算法的高效性。