结合视觉感知与信息量的视点评分方法

随着数字3D内容的不断丰富,对其进行有效展示和索引的需求变得十分迫切,而通过视点评分选择合适的观察视点对解决这一问题具有重要的作用.针对传统的视点评分方法计算时间冗长、结果很难符合人类观察习惯的问题,提出一种基于视觉感知信息量的快速视点评分方法.首先计算相对简单的平均曲率作为代表3D物体视觉特征的要素,随后结合信息熵理论对视点进行评分,使那些能看到尽可能多重要特征且所见特征分布较广的视点分数较高;之后,将这种评分策略应用于最优视点集计算中,利用信息理论对视点集获取信息进行量化,确保利用最少数目的视点有效地认知3D物体.实验结果表明,该方法得到的视点分数及最优视点集质量与目前最好的方法相当,但计算速度更快,无需人工交互.     

基于距离直方图的最优视点选择

基于物体内蕴几何量,提出一种观察三维物体的最优视点选择方法.首先在三维物体表面均匀采样获取采样点,并计算物体形心,然后利用采样点到物体形心的距离来构造距离直方图,最后计算距离直方图的Shannon熵并将其作为衡量视点优劣的标准.根据认知心理学理论,最优视点是存在的,也是恒定的,故文中视点在包围球上选取.实验结果表明,采用该方法获得的最优视点能观察到三维物体更多的功能结构和更显著特征,与其他方法相比更符合人类的感官选择.     

一种具有自终止特性的视点规划方法

提出了一种具有自终止特性的对未知物体模型自动3维测量的视点规划方法。首先在寻找最优视点阶段，利用群矢量链方法获得视觉传感器在下一个视点的观测方向，而视觉传感器在空间的确切位置参数是通过计算矢量场内的边界积分获得，其中获得最大边界积分的空间位置被定为下一个最优视点。同时为了保证对未知物体模型自动3维测量过程具有自终止特性，根据高斯定理，提出由测量所得的点云直接计算出物体体积的方法，并将连续两个视点下物体体积的变化量作为终止测量的依据来判断规划过程是否需要终止或继续。实验结果表明，该具有自终止特性的视点规划方法是可行而且有效的。     

基于视平面上特征计算的视点选择

提出一种能高效观察三维模型的视点选择方法.该方法基于视平面定义一种曲率特征,以度量模型的三维几何特征在视平面上的分布和体现状况,并由此计算与视点相关的一种熵值;然后对各个视点相关的这种熵值进行比较,可得到熵值最大的视点.实验结果表明,采用该方法找到的视点能观察到模型尽可能多的显著特征,与人眼的观察习惯比较接近.与已有方法相比,文中方法计算简单、不需要语义计算、有较高的工作效率.     

基于随机森林的三维人造模型前朝向识别算法

三维模型前朝向的识别是场景合成与重建的基础.针对现有三维模型前朝向识别算法依赖于模型在场景中上下文关系、计算过程复杂等问题,提出一种基于随机森林的三维人造模型前朝向识别算法.首先通过模型简化和计算模型方向包围盒得到候选面;然后基于物体功能设计和心理学等对模型进行形状分析,为候选面提取一组与之相关联的特征;最后利用随机森林理论训练前朝向判别分类器,实现对模型的前朝向识别.实验结果表明,该算法对模型前朝向识别正确率达到80%,能够很好地处理室内场景绝大种类模型,包括目前方法计算错误的模型.     

视点选择Benchmark 的设计与应用

视点选择是图形学的一个重要研究方向,它通过分析模型特征,计算得到符合人类观察习惯并包含更多模型重要信息的视点.近年来,人们提出了很多视点选择算法,然而对这些算法结果的评价仍然停留在感性阶段.设计实现了一个视点选择Benchmark,对视点选择结果进行量化分析:首先,采集30 个测试者对45 个模型的视点选择结果,借以确立真人视点选择的基准视点;接着,以基准视点为参考,分析真人视点选择的一致性和稳定性;最后,应用5 个(共3 类)代表性的算法进行视点选择,并比较其结果与基准视点的差别,得到各视点选择算法的质量及时间效率.实验结果表明,真人视点选择整体具有较好的一致性和稳定性,但对不同特征类型模型,一致性的表现有较大差异;已有视点选择算法对45 个测试模型的结果整体差异不大,相对来说,基于互信息的方法和视平面上特征计算的方法较好,各类视点选择算法对不同特征类型模型结果互有优劣.     

基于分形的水声图像目标探测

针对水声图像中人造物体的探测问题，给出了一种基于分形分析的方法，由于分形模型可以较好地模拟自然物体，而与人工物体存在较大差距，所以以其为主要特征可以准确地将人造物体从自然背景中探测出来。本文讨论了分维的提取方法，根据分形特征将水声图像标记为人造目标区域和非人造目标区域，并对一定噪声干扰下该方法的应用进行了研究，给出了相应的实验结果。实验结果表明，分形特征可以实现人造目标和自然物体的分类，并具有一定的抗噪声性，适宜对水声图像中的目标进行探测和识别。     

基于深度加权法向映射的三维模型检索

提出一种基于深度加权法向映射的三维检索算法，从归一化处理后的物体形状出发，计算关于视点方向的深度加权表面法向统计分布，并将该分布沿视点方向作球面调和分析得到深度加权法向映射特征，通过比较该特征的距离来度量任意三维模型之间的相似性。     

纹理物体表面反射参数的恢复

物体表面反射参数的恢复是新兴的逆向绘制技术的核心问题，提出一种恢复物体表面BRDF模型参数的方法，假定物体表面的BRDF模型可以近似分解为漫射纹理和整体不变的镜面反射分量，则每一点的反射参数可以用Phong模型来近似计算，首先选取物体表面的一片区域，利用该区域在不同视点光照条件下采样的几幅图像，计算得到该区域的平均镜面反射分量，作为整个物体表面的镜面反射分量，然后据此从物体表面各点的采样中分离出漫射分量，计算各点的漫反射参数，得到近似的BRDF模型，实验数据表明算法是有效的。     

基于灭点的单幅图像建模

针对人造物体场景的特点，用人机交互的方法标出物体的平行线和特征点，利用灭点计算出相机的内部参数，进一步计算出旋转矩阵。再根据物体的几何约束关系，得到平移向量和物体模型的空间信息。然后从图像中提取纹理，经过变换后，映射到模型上去，使之具有相片般的真实感，并能够进行实时漫游。     

Automatic upright orientation and good view recognition for 3D man-made models

Many graphics applications require that 3D models automatically be presented upright and from a good view. We propose a method that simultaneously recognizes upright orientation and good view for 3D man-made models. The strategy is to determine the best base on which a 3D model can stand upright from a small set of candidate bases. Every candidate base is composed of clustered facets of the simplified convex hull of the given 3D model. Next, a proposed UV-measurement selects the best base from the candidate bases using weighted feature-based evaluation functions based on geometrical, physical, and visual aspects. Our method has been tested using a public 3D model database and compared with previous methods. As experimental results show, our method outperforms previous work in both efficiency and accuracy.     

Upright orientation of 3D shapes with Convolutional Networks

Posing objects in their upright orientations is the very first step of 3D shape analysis. However, 3D models in existing repositories may be far from their right orientations due to various reasons. In this paper, we present a data-driven method for 3D object upright orientation estimation using 3D Convolutional Networks (ConvNets), and the method is designed in the style of divide-and-conquer due to the interference effect. Thanks to the public big 3D datasets and the feature learning ability of ConvNets, our method can handle not only man-made objects but also natural ones. Besides, without any regularity assumptions, our method can deal with asymmetric and several other failure cases of existing approaches. Furthermore, a distance based clustering technique is proposed to reduce the memory cost and a test-time augmentation procedure is used to improve the accuracy. Its efficiency and effectiveness are demonstrated in the experimental results.     

Unsupervised upright orientation of man-made models

Most man-made models can be posed at a unique upright orientation which is consistent to human sense. However, since produced by various techniques, digital man-made models, such as polygon meshes, might be sloped far from the upright orientation. We present a novel unsupervised approach for finding the upright orientation of man-made models by using a low-rank matrix theorem based technique. We propose that projections of the models could be regarded as low-rank matrices when they have been posed at axis-aligned orientations. The models are to be iteratively rotated by using the recently presented TILT technique, in order to ensure that their projections have optimal low-rank observations. After that, the upright orientation can be easily picked up from the six axis-aligned candidate orientations by analysis on geometric properties of the model. The approach does not require any other training set of models and should be regardless of the model quality. A number of experiments will be shown to illustrate the effectiveness and robustness of the proposed approach.     

目标不同视角下观察信息的迁移和再利用

基于视觉信息的目标检测和识别模型在训练时往往依赖于来自于训练样本的视角信息,然而附带了视角信息的训练样本通常只有很少的数据库可以提供。当此类信息缺失时,传统的通用目标检测系统通常通过一些非监督学习方法来对样本的视角信息进行粗略估计。本文改进并引入了一种选择性迁移学习方法即TransferBoost方法来解决目标视角信息缺失的问题。本文TransferBoost方法基于GentleBoost框架实现,该方法通过重新利用其它类别样本中的先验信息来提升当前类别样本的学习质量。当给定一个标定完善的样本集作为源数据库时,TransferBoost通过同时调整每个样本的权值和每个源任务的权值实现样本级和任务级的两级知识迁移。这种双层迁移学习更有效地从混合了相关源数据和不相关源数据的数据集中提取了有用的信息。实验结果表明,和直接使用传统的机器学习方法相比较,迁移学习方法所需要的训练样本数大大减少,从而降低了目标检测与识别系统的训练代价,扩展了现有系统的应用范围。     

Category‐Specific Salient View Selection via Deep Convolutional Neural Networks

In this paper, we present a new framework to determine up front orientations and detect salient views of 3D models. The salient viewpoint to human preferences is the most informative projection with correct upright orientation. Our method utilizes two Convolutional Neural Network (CNN) architectures to encode category‐specific information learnt from a large number of 3D shapes and 2D images on the web. Using the first CNN model with 3D voxel data, we generate a CNN shape feature to decide natural upright orientation of 3D objects. Once a 3D model is upright‐aligned, the front projection and salient views are scored by category recognition using the second CNN model. The second CNN is trained over popular photo collections from internet users. In order to model comfortable viewing angles of 3D models, a category‐dependent prior is also learnt from the users. Our approach effectively combines category‐specific scores and classical evaluations to produce a data‐driven viewpoint saliency map. The best viewpoints from the method are quantitatively and qualitatively validated with more than 100 objects from 20 categories. Our thumbnail images of 3D models are the most favoured among those from different approaches.     

Motion estimation by decoupling rotation and translation in catadioptric vision

Previous works have shown that catadioptric systems are particularly suited for egomotion estimation thanks to their large field of view and thus numerous algorithms have already been proposed in the literature to estimate the motion. In this paper, we present a method for estimating six degrees of freedom camera motions from central catadioptric images in man-made environments. State-of-the-art methods can obtain very impressive results. However, our proposed system provides two strong advantages over the existing methods: first, it can implicitly handle the difficulty of planar/non-planar scenes, and second, it is computationally much less expensive. The only assumption deals with the presence of parallel straight lines which is reasonable in a man-made environment. More precisely, we estimate the motion by decoupling the rotation and the translation. The rotation is computed by an efficient algorithm based on the detection of dominant bundles of parallel catadioptric lines and the translation is calculated from a robust 2-point algorithm. We also show that the line-based approach allows to estimate the absolute attitude (roll and pitch angles) at each frame, without error accumulation. The efficiency of our approach has been validated by experiments in both indoor and outdoor environments and also by comparison with other existing methods.     

Efficient view selection by measuring proxy information

View selection aims at finding good views that can watch model information as much as possible. However, existing view evaluation methods used to measure model properties are generally complex and very time‐consuming. In this paper, we address this challenge by generating succinct proxy information to measure the view quality. The proxy information is generated by principal component analysis of the model to obtain its six principal viewing directions and consists of the representative information computed by the principal viewing directions. We discuss and validate the effectiveness of the proxy information for view evaluation. Thus, measuring the proxy information for view evaluation, the time complexity can be reduced as it is independent of the facet number of the model. This is superior to existing methods by straight measurement of model properties and provides acceleration for view selection. Experimental results show that we can obtain good views as state‐of‐the‐art methods and speed up view selection by at least two orders of magnitudes and achieve more acceleration when the models have more facets. Copyright © 2016 John Wiley & Sons, Ltd.     

多点测试的多模型机动目标跟踪算法

针对机动目标跟踪问题,提出了一种多点测试多模型粒子滤波算法(Independence multi-try method, IMTM).整个算法分为两个阶段,第一阶段为利用多点测试(Multi-try method, MTM)结构从各模型产生的粒子中选取一个最优粒子,实现了模型间的交互;第二阶段为利用IMH (Independence Metropolis-Hastings)滤波算法对第一阶段产生的粒子进行取舍,完成整个状态估计.相对于传统的交互式多模型(Interacting multiple model, IMM)算法,该算法无需事先设定模型转移概率 矩阵且为整体并行结构,结构简单,能够充分地交互各模型之间的粒子,进而自动有效地调整各模型权值比重,降低了人为干扰.仿真表明,该算法能够有效地降低滤波峰值误差,整体跟踪精度较高,算法的实时性较好.     

Visual saliency guided textured model simplification

Mesh geometry can be used to model both object shape and details. If texture maps are involved, it is common to let mesh geometry mainly model object shapes and let the texture maps model the most object details, optimising data size and complexity of an object. To support efficient object rendering and transmission, model simplification can be applied to reduce the modelling data. However, existing methods do not well consider how object features are jointly represented by mesh geometry and texture maps, having problems in identifying and preserving important features for simplified objects. To address this, we propose a visual saliency detection method for simplifying textured 3D models. We produce good simplification results by jointly processing mesh geometry and texture map to produce a unified saliency map for identifying visually important object features. Results show that our method offers a better object rendering quality than existing methods.