三维重建中的剖视图处理

剖视图处理能够有助于更好地理解视图,提高三维重建的效率。基于对剖视图自身特点的分析,提出了三维重建中对剖视图的处理算法,即通过剖视图信息的智能提取生成基元体,将基元体同由视图轮廓生成的参照体进行布尔运算,从而得到最终的目标体。算法能够处理包括全剖视图、半剖视图以及平行剖视图在内的剖视图,最后给出了应用算例以验证算法的可行性。     

“三维视觉—语言”推理技术的前沿研究与最新趋势

三维视觉推理的核心思想是对点云场景中的视觉主体间的关系进行理解。非专业用户难以向计算机传达自己的意图,从而限制了该技术的普及与推广。为此,研究人员以自然语言作为语义背景和查询条件反映用户意图,进而与点云的信息进行交互以完成相应的任务。此种范式称做“三维视觉—语言”推理,在自动驾驶、机器人导航以及人机交互等众多领域广泛应用,已经成为计算机视觉领域中备受瞩目的研究方向。过去几年间,“三维视觉—语言”推理技术迅猛发展,呈现出百花齐放的趋势,但是目前依然缺乏对最新研究进展的全面总结。本文聚焦于两类最具代表性的研究工作,锚框预测和内容生成类的“三维视觉—语言”推理技术,系统性概括领域内研究的最新进展。首先,本文总结了“三维视觉—语言”推理的问题定义和现存挑战,同时概述了一些常见的骨干网络。其次,本文按照方法所关注的下游场景,对两类“三维视觉—语言”推理技术做了进一步细分,并深入探讨了各方法的优缺点。接下来,本文对比分析了各类方法在不同基准数据集上的性能。最后,本文展望了“三维视觉—语言”推理技术的未来发展前景,以期促进该领域的深入研究与广泛应用。     

视觉知识：跨媒体智能进化的新支点

回顾跨媒体智能的发展历程,分析跨媒体智能的新趋势与现实瓶颈,展望跨媒体智能的未来前景。跨媒体智能旨在融合多来源、多模态数据,并试图利用不同媒体数据间的关系进行高层次语义理解与逻辑推理。现有跨媒体算法主要遵循了单媒体表达到多媒体融合的范式,其中特征学习与逻辑推理两个过程相对割裂,无法综合多源多层次的语义信息以获得统一特征,阻碍了推理和学习过程的相互促进和修正。这类范式缺乏显式知识积累与多级结构理解的过程,同时限制了模型可信度与鲁棒性。在这样的背景下,本文转向一种新的智能表达方式——视觉知识。以视觉知识驱动的跨媒体智能具有多层次建模和知识推理的特点,并易于进行视觉操作与重建。本文介绍了视觉知识的3个基本要素,即视觉概念、视觉关系和视觉推理,并对每个要素展开详细讨论与分析。视觉知识有助于实现数据与知识驱动的统一框架,学习可归因可溯源的结构化表达,推动跨媒体知识关联与智能推理。视觉知识具有强大的知识抽象表达能力和多重知识互补能力,为跨媒体智能进化提供了新的有力支点。     

混合增强视觉认知架构及其关键技术进展

智能视觉系统虽然在大规模信息的特征检测、提取与匹配等处理上具备一定优势，但是在深层次认知上仍存在不确定性和脆弱性，尤其是针对视觉感知基础上的视觉认知任务，相关数理逻辑和图像处理方法并未实现质的突破，智能算法难以取代人类执行较为复杂的理解、推理、决策和学习等操作。为助力智能视觉感知和认知技术的进一步发展，本文总结了混合增强智能在视觉认知领域的应用现状，给出了混合增强视觉认知的基本架构，并对可纳入该架构下的应用领域及关键技术进行了综述。首先，在分析智能视觉感知内涵和基本范畴的基础上，融合人的视觉感知与心理认知，探讨混合增强视觉认知的定义、范畴及其深化过程，对不同的视觉信息处理阶段进行对比，进而在分析相关认知模型发展现状的基础上，构建混合增强视觉认知的基本框架。该架构不仅可依靠智能算法进行快速地检测、识别、理解等处理，最大限度地挖掘"机"的计算潜能，而且可凭借适时、适当的人工推理、预测和决策有效增强系统认知的准确性和可靠性，最大程度地发挥人的认知优势。其次，分别从混合增强的视觉监测、视觉驾驶、视觉决策以及视觉共享等4个领域探讨可纳入该架构的代表性应用及存在的问题，指出混合增强视觉认知架构是现有技术条件下能够更好地发挥计算机效能、减轻人处理信息压力的方式。最后，基于高、中、低计算机视觉处理技术体系，分析混合增强视觉认知架构中部分中高级视觉处理技术的宏观、微观关系，重点综述可视化分析、视觉增强、视觉注意、视觉理解、视觉推理、交互式学习以及认知评估等关键技术。混合增强视觉认知架构有助于突破当前视觉信息认知"弱人工智能"的瓶颈，将有力促进智能视觉系统向人机深度融合方向发展。下一步，还需在纯粹的基础创新、高效的人机交互、柔性的连接通路等方面开展更加深入的研究。     

基于双枝模糊集的一致性模糊变权Petri网攻击模型

以Petri网为基础,结合双枝模糊集理论,定义了一种新型的一致性网络攻击模型CBBVFPN,并根据CBBVFPN模型的特点构造出了相应的形式化推理算法。该模型弥补了以往基于Petri网的攻击模型共同具有的一些缺陷,同时对“AND”和“OR”两种基本Petri结构进行了扩展。在推理过程中,为了防止以一个真实度很低的命题为前提继续推理,得出真实度更低的命题而背离事实,提出规则演化和变权的方法来处理真实度低的命题。推理算法使用统一的方式表示肯定命题和否定命题,并且在算法中加入控制条件,以保证推理过程和推理结果的一致性。最后以Botnet攻击为实例验证了本算法。     

网络视觉数据语义关联理解研究综述

针对现有大多数视觉数据理解孤立地处理单个视觉对象且专注于其固有特性,对网络异构分布及相互关联等特性重视不够,导致出现求解困难、计算效率低、仅能理解低层语义等问题,阐述面向大规模网络视觉数据的关联推理与语义理解研究热点,分析国内外研究现状,对该方向的发展趋势进行展望。     

跨模态视觉问答与推理研究进展

随着社交媒体和人机交互技术的快速发展,视频、图像以及文本等多模态数据在互联网中呈爆炸式增长,因此多模态智能研究受到关注。其中,视觉问答与推理任务是跨模态智能研究的一个重要组成部分,也是人类实现人工智能的重要基础,已成功应用于人机交互、智能医疗以及无人驾驶等领域。本文对视觉问答与推理的相关算法进行了全面概括和归类分析。首先,介绍了视觉问答与推理的定义,并简述了当前该任务面临的挑战;其次,从基于注意力机制、基于图网络、基于预训练、基于外部知识库和基于可解释推理机制5个方面对现有方法进行总结和归纳;然后,全面介绍了视觉问答与推理常用公开数据集,并对相关数据集上的已有算法进行详细分析;最后,对视觉问答与推理任务的未来方向进行了展望。     

使用元知识检查知识相容性和加速规则动态查找

本文提出一种基于对规则语义“了解”的元知识对知识进行处理的方法:利用元知识静态检查知识的相容性,并利用元知识对规则进行静态处理,建立基于逻辑依赖关系的半序集,加快推理过程中每个推理周期可用规则的查找速度.并给出了有关的算法.     

可由用户持续发展的几何自动推理平台的推理算法

目前的几何定理证明器都不具有可持续性。提出一种结构具有一般性的知识表示和能够统一处理所有规则的推理算法，初步实现了可由用户持续发展的几何自动推理平台。该推理平台允许用户添加几何知识，如几何对象、谓词和规则，并可以综合使用多种推理算法，如前推搜索法和一部分面积法，它将更适合用于几何教学。     

概率图模型推理方法的研究进展

近年来概率图模型已成为不确定性推理的研究热点,在人工智能、机器学习与计算机视觉等领域有广阔的应用前景.根据网络结构与查询问题类型的不同,系统地综述了概率图模型的推理算法.首先讨论了贝叶斯网络与马尔可夫网络中解决概率查询问题的精确推理算法与近似推理算法,其中主要介绍精确推理中的VE算法、递归约束算法和团树算法,以及近似推理中的变分近似推理和抽样近似推理算法,并给出了解决MAP查询问题的常用推理算法;然后分别针对混合网络的连续与混合情况阐述其推理算法,并分析了暂态网络的精确推理、近似推理以及混合情况下的推理;最后指出了概率图模型推理方法未来的研究方向.     

Solid reconstruction from orthographic opaque views using incremental extrusion

The reconstruction of 3-D solids from 2-D projections is an important research topic in reverse engineering. The reconstruction can be grouped into two categories: single-view approach and multiple-view approach. Each approach can be classified as wireframe, BRep or CSG. However, not many CSG approaches have been reported in the literature. The methods are also restricted to uniform-thickness objects or require user interaction. The method proposed in this paper employs the CSG approach. A 3-D solid computer model is reconstructed from 2-D line drawings of six orthographic opaque views, viz. top, front, left, right, bottom and rear views. Firstly, the six views are grouped into three pairs. For each pair of views, segmented areas from one of the two views (called g-view) is incrementally extruded according to the information in the neighbouring view (called d-view). Extrusion primitive solids are generated during the incremental extrusion. All primitive solids are then unioned into an extrusion solid. Finally, all extrusion solids are intersected to give a unique 3-D solution object.     

Creation and boundary evaluation of CSG-models

This paper deals with one particular type of solid model representation. Constructive solid geometry (CSG) describes solids in terms of primitive shapes combined by Boolean operators. The goal of this paper is to present an algorithm suitable for the visualization of a given CSG-object for verification purposes. To accomplish this, the CSG representation is to be converted into a polygonal boundary representation (B-Rep), which allows the easy display of the solid's visible edges and surfaces. An algorithm performing this conversion is called a boundary evaluator. This paper describes the fundamental concepts leading to this new algorithm and presents some examples.The boundary evaluator described here follows the following steps: (1) A polygonal mesh is created that covers the primitive shapes. (2) Interfering polygons are repertedly subdivided until the resulting mesh contains the intersection curves between the primitives. (3) The location and validity of all polygons must be determined since only valid polygons reflect the solid's final shape. (4) Adjoining coplanar polygons are collected and merged by deleting their common edges. The final result of this procedure is a consistent, nonredundant polygon mesh describing the given solid model. Now the object may be displayed by simple perspective plots, by isometric projections, by hidden line drawings, or through the use of hidden surface removal techniques as color-shaded pictures. The algorithm presented here has been successfully used to create a variety of renderings of various CSG-models. The interpretation, display, and validation of computer-generated CSG models is essential for the successful use of such models in the engineering design process.     

Embedding visual cognition in 3D reconstruction from multi-view engineering drawings

In this paper we discuss how to reconstruct 3D models from multi-view engineering draws by employing human engineers’ approaches. Human's ‘divide and conquer’ interpretation strategy in visual cognition is simulated, and successfully carried out on the basis of spatial division of 3D object space. At first, a volume-oriented method is utilized to decompose the 3D object space in a set of 3D ‘cell-boxes’ whose three-view bounding rectangles will isolate its related sub-projections from input projections views, and in every cell-box a cell primitive is implied. Then, a 3D model of each cell primitive is ‘locally’ generated from its ‘sub-projection’ views by wire-frame oriented algorithms. The final interpretation result of the overall projection-views is a ‘Union’ of these cell primitives. To deal with the ambiguity, a visual reasoning engine is implemented on the basis of principles from Gestalt psychology. It will be activated to pick out the most reasonable interpretation, when ambiguities are generated during the reconstruction process. The section views are also incorporated in getting rid of the ambiguity. Moreover, we design a natural and convenient interaction way to encourage the user to be involved in the process while interpreting complex projections. The key steps of this human-like reconstruction approach are presented in detail.     

Recovering solid geometric object from single line drawing image

Many educational materials contain a lot of solid geometric figures. The solid geometric objects in these figures are usually drawn as 2D line drawings thus have lost their 3D information. This paper presents a method to recover the 3D information of the solid geometric object from single line drawing image taken from the geometric books, which would be used to help the users better present and understand the solid geometric object on their mobile devices. The main advantage of our method is the abilitYTo handle inaccurately processed sketches as opposed to the previous methods which require perfect line drawings as inputs. Our method consists of three main steps as follows. First, the sketch of the input line drawing image is automatically extracted and further represented as an undirected graph. Second, candidate 3D models from the pre-built 3D model database are found by graph similarity-based searching and sub-graph isomorphism matching. Third, for each candidate 3D model, the model parameters, the rotation and the translation aligning the model with the sketch are found by minimizing an objective function which is composed of the residuals between the vertices of the sketch and the 2D projections of the candidate model’s vertices, and an optimal reconstruction solution is further selected as the final result. Extensive experimental results demonstrate the effectiveness and robustness of our method for recovering the solid geometric object from single line drawing image.     

Boolean operations with implicit and parametric representation of primitives using R-functions

We present a new and efficient algorithm to accurately polygonize an implicit surface generated by multiple Boolean operations with globally deformed primitives. Our algorithm is special in the sense that it can be applied to objects with both an implicit and a parametric representation, such as superquadrics, supershapes, and Dupin cyclides. The input is a constructive solid geometry tree (CSG tree) that contains the Boolean operations, the parameters of the primitives, and the global deformations. At each node of the CSG tree, the implicit formulations of the subtrees are used to quickly determine the parts to be transmitted to the parent node, while the primitives' parametric definition are used to refine an intermediary mesh around the intersection curves. The output is both an implicit equation and a mesh representing its solution. For the resulting object, an implicit equation with guaranteed differential properties is obtained by simple combinations of the primitives' implicit equations using R-functions. Depending on the chosen R-function, this equation is continuous and can be differentiable everywhere. The primitives' parametric representations are used to directly polygonize the resulting surface by generating vertices that belong exactly to the zero-set of the resulting implicit equation. The proposed approach has many potential applications, ranging from mechanical engineering to shape recognition and data compression. Examples of complex objects are presented and commented on to show the potential of our approach for shape modeling.     

Octree creation via C.S.G. definition

The common method for generating the octrees of complex objects, is based upon generating the octrees of several pre-defined primitives and applying Boolean operations on them. Regardless how the octrees representing the primitives are generated (top-down or bottom-up) the octree of a desired object is obtained by performing Boolean operations among the primitives comprising the object according to the object's CSG (constructive solid Geometry) representation. When carrying out this procedure, most of the computing and memory resources are used for generating and storing the octants comprising the primitives. However, the majority of those octants are not required for the representation of the final object. In this paper the extention of the top-down approach to the CSG level (i.e., generating the octree of an object directly from its CSG representation) is proposed. With this method there is no need to generate the octrees of the primitives comprising the object nor to perform Boolean operations on them. Moreover, only these octants which belong to the final object are generated.     

基于扫描跟踪元的快速碰撞检测

针对扫描实体在碰撞检测中的应用,提出一种基于扫描跟踪元的快速碰撞检测算法.首先根据运动模型上的采样点构造扫描元曲线群;然后沿着运动路径建立截平面,通过截平面与扫描元曲线群的交点构建特殊平面网格,提取出该平面网格的外轮廓顶点集;最后通过外轮廓点集构造扫描跟踪元.通过扫描跟踪元与环境的相交性检测,能快速、有效地检测出碰撞区域和碰撞时间.实验结果表明,该算法能在保证高精度检测的前提下有效地减少计算量,提高工作效率,因而具有较高的应用价值.     

面向内容的工程图识别与理解综述

对工程图识别与理解研究现状进行深入和系统的综述。工程图识别与理解 的核心在于利用知识表示、图形匹配、符号识别、几何推理、语义提取等技术,自动获取工 程图中各种显式描述（如几何图元、工程符号、工程对象等）及隐式信息（如设计语义等）。 分别对工程图矢量化、工程符号识别及工程对象识别进行了系统分析。在此基础上对工程图 识别性能评测、基准库等进行总结。最后对现有研究的难点及进一步研究方向进行了展望。     

Reconstruction of 3D interacting solids of revolution from 2D orthographic views

3D CAD is replacing 2D CAD to improve efficiency of product design and manufacturing. Therefore, converting legacy 2D drawings into 3D solid models is required. CSG based approaches reconstruct solid models from orthographic views more efficiently than traditional B-rep based approaches. A major limitation of CSG based approaches has been the limited domain of objects that can be handled. This paper aims at extending the capabilities of CSG based approaches by proposing a hint-based recognition of interacting solids of revolution. This approach can handle interacting solids of revolution as well as isolated solids of revolution.     

基于参数化的系列化图形尺寸自动标注方法

为了提高系列化图形的设计效率,降低工程制图的尺寸标注的复杂度,结合二三维一体化的设计思想,研究系列化图形的尺寸标注,提出并实践了一种系列化图形的自动标注方法--依据参数化进行实体的三维造型,然后将其进行投影生成二维视图,在二维视图上进行尺寸交互标注.研究了视图与尺寸标注之间的联系,提取关键信息,利用该关键信息对后续生成的系列化图形进行尺寸标注.最后以实例展示了自动标注的过程,表明了该方法的可行性,有效性.