基于ASODVS的全景点云数据获取技术的研究

针对现有三维激光扫描设备不能同时获取空间点云的坐标信息和色彩信息,多视角点云数据配准复杂等问题,采用了一种将全方位视觉传感器ODVS(Omni-directional Vision Sensor)和可移动360°面激光光源相结合的主动式全景视觉传感器ASODVS(Active Stereo Omni-directional Vision Sensor)来获取点云数据。通过人机接口对ASODVS的扫描步长和速度等进行控制,实现了边扫描边获取全景点云数据;对获取点云数据空间信息过程中的关键技术——激光投射点提取算法进行了深入研究;采用了3种不同的全景激光线提取算法对全景切片图像中激光投射点进行提取,并实验研究了各自的优劣;实验结果表明,对于ASODVS而言,帧间差提取算法能有效快速准确的提取三维全景场景中激光投射点,并具有消耗计算机资源少、操作过程自动化、生成数据有序等优点。     

基于VTK的三维点云曲面重建研究

针对三维点云数据重建效率低、不能实时交互等问题,利用鲁棒性强的Power Crust算法和三维可视化类库Visualization Toolkit (VTK)的良好并行机制与强大的图像处理能力,实现了三维点云数据曲面快速重建.该算法使用Power Crust对三维点云进行曲面重建,接着对得到的网格进行线性调整、简化和平滑,最后引入VTK进行渲染、绘制、显示,并实时交互.实验结果表明,该算法可以加快散乱点云数据的重建速度,较好地保持了点云数据的拓扑结构,提高了曲面重建的精确性和鲁棒性,且交互性强,适合实时处理.     

面向三维模型分割的边界感知点云神经网络

为了能够更好地应用深度神经网络学习三维模型的空间特征,获得更好的三维模型分割效果,提出面向三维模型分割的边界感知点云神经网络.首先,采用边界感知的网格点云化方法,将网格分割问题转化成点云标记问题;然后,利用数据切片方法对转化而来的点云数据进行重采样;最后,利用不同大小卷积核的滤波器提取点云数据的空间特征,并将点云标记的结果对应到原网格模型,得到三维模型分割的结果.在ShapeNetCore数据库上的实验结果表明,该方法不仅能够明显地提高分割的准确率,而且具有边界感知的特性,能够有效地避免过分割现象.     

基于三维激光雷达的道路边界提取和障碍物检测算法

为了快速有效地提取智能车辆在不同环境下的道路环境信息,提出基于三维激光雷达的道路边界提取和障碍物检测算法.首先,对三维激光雷达点云数据进行栅格化滤波处理,利用单束激光点云空间邻域联合分割的方法进行空间分析,得到点云平滑度特征图像.然后,采用自适应方向搜索算法获取道路边界候选点,并进行聚类分析和曲线拟合.最后,对道路边界约束下可通行区域内点云进行聚类分割,获得道路内障碍物方位和距离信息.实验表明,文中算法能够实时准确地提取道路边界和障碍物位置信息,满足智能车环境建模和路径规划的需求.     

自动驾驶汽车虚拟测试中的树木点云生成方法

自动驾驶汽车虚拟测试已成为自动驾驶或车路协同测试评价的一个重要手段,三维激光雷达数据模拟生成是自动驾驶汽车虚拟测试中的重要任务之一,目前多采用基于飞行时间原理的几何模型方法生成激光雷达三维点云数据,该方法生成点云实时性较差。布告牌是虚拟场景中常采用的树木建模方法,由于布告牌仅由两个矩形面片即八个三角形面片组成,直接采用布告牌方法生成的三维点云数据难以反映树木的真实空间信息。针对上述问题,提出了一种基于布告牌空间变换的快速树木三维点云生成方法。以布告牌的纹理图像为依据,根据纹理透明度获取树木二维平面点云分布,经二维树木点云的轮廓提取,结合树木结构的先验知识进行旋转、随机偏移和尺度变换,以更少的三角形面片数和更小的计算代价获得树木的三维点云数据。提出了一种空间直方图三维点云相似度评价方法,将三维点云空间量化为若干个子空间,获得三维点云的投影空间直方图,采用巴氏系数计算投影空间直方图相似度,以投影空间直方图加权相似度作为点云相似度评价值。实验结果表明,基于布告牌空间变换方法和几何模型方法生成的云杉等三种树木的三维点云数据的平均相似度在90%以上,且该方法生成树木点云的时间仅是几何模型法的1%,因此布告牌空间变换树木三维点云生成方法快速且准确,可以满足自动驾驶汽车虚拟测试的性能要求。     

基于结构光和极线约束的三维重建

结构光可以简单、快速并且精确地对被测物体进行三维信息提取,从而得到三维数据点云.而在结构光三维信息获取系统中,条纹的准确检测是影响系统精度的关键因素之一.为了改善这一问题,先利用格雷码对图像条纹进行粗定位,再结合相位码准确地对图像务纹进行细定位并利用极线约束把图像匹配过程中的二维搜索降为一维搜索.结构光条纹和极线的交点就是图像的匹配点,对匹配点空间前方交会即可得到物体的三维点云.反复试验表明,该方法能取得较好的效果.     

基于分层特征化网络的三维人脸识别

针对三维人脸多表情、多姿态变化同时存在,人脸点云数据不同程度缺失的问题,探索性地将三维点云人脸数据应用于PointNet系列的分类网络并进行了识别结果的对比与分析,然后提出了一种新的网络框架——HFN。首先,在数据预处理后随机采样固定点数的点云;其次,将固定点数的人脸点云输入SA模块,以获取局部区域的质心点、邻域点并提取局部区域的特征,然后拼接由DSA模块基于多方向卷积提取的点云空间结构特征;最后,利用全连接层进行三维人脸的分类,从而实现三维人脸识别。在CASIA数据库上的结果显示,所提方法的平均识别率为96.34%,优于PointNet、PointNet++、PointCNN和空间聚合网络（SAN）这几种分类网络。     

集成RGB-D语义分割网络的室内语义地图构建

针对传统视觉SLAM准确度低、实时性差、缺乏语义的问题,提出一种全新的RGB-D语义分割网络,利用室内场景中受光照等条件影响较小的深度信息来提高分割的准确性,并且设计了轻量级多尺度残差模块(MRAM)和空间金字塔池化模块(ASPP)来轻量化分割网络、提高分割的精度。首先输入的图像序列进入ORB-SLAM2网络进行关键帧筛选,之后关键帧送入语义分割网络得到二维语义标签,再将二维语义信息映射到三维点云空间,最后使用贝叶斯算法更新三维地图得到全局一致的三维点云语义地图。实验采用NYUv2数据集验证语义分割网络性能,采用TUM数据集构建点云语义地图,结果表明,提出的语义分割网络性能和速度优于现有的模型,且此语义分割网络与视觉SLAM相结合可以满足高精度、实时的稠密三维语义点云地图构建要求。     

基于分层特征化网络的三维人脸识别

针对三维人脸多表情、多姿态变化同时存在，人脸点云数据不同程度缺失的问题，探索性地将三维点云人脸数据应用于PointNet系列的分类网络并进行了识别结果的对比与分析，然后提出了一种新的网络框架——HFN。首先，在数据预处理后随机采样固定点数的点云；其次，将固定点数的人脸点云输入SA模块，以获取局部区域的质心点、邻域点并提取局部区域的特征，然后拼接由DSA模块基于多方向卷积提取的点云空间结构特征；最后，利用全连接层进行三维人脸的分类，从而实现三维人脸识别。在CASIA数据库上的结果显示，所提方法的平均识别率为96.34%，优于PointNet、PointNet++、PointCNN和空间聚合网络（SAN）这几种分类网络。     

基于注意力机制的三维点云车辆目标检测

针对自动驾驶场景下三维点云车辆的识别和定位问题,提出了一种基于注意力机制的三维点云车辆目标检测算法.算法将稀疏无序的点云空间划分成等距规则的体素表示,用三维稀疏卷积和辅助网络同步从所有体素中提取内部点云特征,进而生成鸟瞰图.但在将内部三维的点云特征转化为二维的鸟瞰图后,通常会造成目标空间特征信息丢失,使得最终检测结果以及方向性预估差.为进一步提取鸟瞰图中特征信息,提出了一种注意力机制模块,其中包含两种注意力模型,并对其采用首、中、尾的"立体式"布局结构,实现对鸟瞰图中特征信息的放大和抑制,最后使用卷积神经网络和PS-Warp变换机制对处理过后的鸟瞰图进行三维目标检测.实验表明,该算法在保证实时检测效率的前提下,与现有算法相比,具有更好的方向预估性以及更高的检测精度.     

Sensor-based motion planning in three dimensions for a highly redundant snake robot

A strategy is described for real-time motion planning for a highly redundant snake-like robot manipulator operating in a three-dimensional (3D) environment filled with unknown obstacles of arbitrary shape. The robot consists of many (say 30 or 50) links serially connected by universal joints (such a joint allows a 3D rotation of one link relative to the other). The robot's sensors allow it to sense objects in the vicinity of any points of its body. The task is to move the robot's tip point (its head) from its starting position to a specified target position, collision-free for the whole robot's body. To achieve the efficiency necessary for real-time computation, an iterative procedure is proposed which makes use of a unit motion for a single link based on the tractrix curve. This choice also results in automatically achieving a motion that is 'natural' (in that the joint displacements tend to 'die out' in the direction from head to tail) as well as locally optimal.     

Online tracking and mimicking of human movements by a humanoid robot

This paper describes a humanoid robot system that can capture and mimic the motion of human body parts in real-time. The underlying vision system is able to automatically detect and track human body parts such as hands and faces in images captured by the robot's eyes. It is based on a probabilistic approach that can detect and track multiple blobs in a 60-Hz stereo image stream on a standard dual processor PC. A random jerk model is employed to approximate the observed human motion and a Kalman filter is used to estimate its parameters (three-dimensional positions, velocities and accelerations). This enables the system to realistically mimic the perceived motion in real-time.     

变电站三维模型可视化交互研究

随着电网信息化建设的加快,通过对变电站主设备和辅助设备进行三维场景建模,实现变电站场景的数字化还原,并根据变电站内设备信息将三维节点与数据库点进行逻辑关联,将变电站监控推向数字化、可视化、智能化的全新高度。本文旨在研究一种通用的三维交互方法,通过规范化三维模型的动态节点命名的方法,实现三维模型和三维程序之间的交互信息的自动化提取和控制,从而对变电站主设备监测数据、辅助设备监测数据等进行了全面监视,构建智慧变电站三维可视化交互平台,更加高效直观地全面监视变电站,增强设备的管控力。     

Automatic head and facial feature extraction based on geometry variations

Facial anthropometry plays an important role in ergonomic applications. Most ergonomically designed products depend on stable and accurate human body measurement data. Our research automatically identifies human facial features based on three-dimensional geometric relationships, revealing a total of 67 feature points and 24 feature lines — more than the definitions associated with MPEG-4. In this study, we also verify the replicability, robustness, and accuracy of this feature set. Even with a lower-density point cloud from a non-dedicated head scanner, this method can provide robust results, with 86.6% validity in the 5 mm range. We also analyze the main 31 feature points on the human face, with 96.7% validity of less than 5 mm.     

Automatic alignment for virtual fitting using 3D garment stretching and human body relocation

In recent years, researches on virtual fitting are mostly based on the data of garment pieces through Computer Aided Design. However, three-dimensional garment models are more intuitive, integrated and easy to access and manage. Therefore, we propose and design a method to automatically align virtual humans with three-dimensional garment models. Firstly, we basically align two models through three-dimensional geometric calculation and postural alteration of virtual humans. Then, we detect the penetration phenomena and define the points of the garment model penetrating into the human model as penetration points. Based on the penetration detection, we use Poisson mesh deformation to adjust the penetration points by imitating the adjustment process of dragging clothes. Finally, the results obtained from alignment can be directly used for physical-based simulation. Experimental results show that the method we propose can achieve good effects of alignment, and has good versatility and practicality.     

Classification of lidar bare-earth points,buildings, vegetation,and small objects based on region growing and angular classifier

In recent years, light detection and ranging (lidar) systems have been intensively used in different urban applications such as map updating, communication analysis, virtual city modelling, risk assessment, and monitoring. A prerequisite to enhance lidar data content is to differentiate ground (bare earth) points that yield digital terrain models and off-terrain points in order to classify urban objects and vegetation. The increasing demand for a fast and efficient algorithm to extract three-dimensional urban features was the motive behind this work. A new combined approach to extract bare-earth points is proposed, and a novel methodology to automatically classify airborne laser data into different objects in an urban area is presented. In addition, a new concept of angular classification is introduced to differentiate buildings from vegetation and other small objects. The new angular classifier analyses the distribution of bare-earth points around unclassified point clusters to determine whether a cluster can be classified either as building or as vegetation. The experimental results confirm the high accuracy achieved to automatically classify urban objects in flat complex areas.     

Automatic sensor placement from vision task requirements

The problem of automatically generating the possible camera locations for observing an object is defined, and an approach to its solution is presented. The approach, which uses models of the object and the camera, is based on meeting the requirements that: the spatial resolution be above a minimum value, all surface points be in focus, all surfaces lie within the sensor field of view and no surface points be occluded. The approach converts each sensing requirement into a geometric constraint on the sensor location, from which the three-dimensional region of viewpoints that satisfies that constraint is computed. The intersection of these regions is the space where a sensor may be located. The extension of this approach to laser-scanner range sensors is also described. Examples illustrate the resolution, focus, and field-of-view constraints for two vision tasks     

iFIX组态软件中通用趋势曲线的设计与实现

介绍了一种iFIX环境下适用于各类监控系统的通用趋势曲线设计,通过ADO技术访问历史和实时数据库,自动获取iFIX历史定义的标签点作为曲线数据点,并且解决了在趋势曲线中标签点中文化的问题,可以方便地移植到新建监控系统的组态工程中,有效地降低了开发难度。     

A Hierarchical RBF Online Learning Algorithm for Real-Time 3-D Scanner

In this paper, a novel real-time online network model is presented. It is derived from the hierarchical radial basis function (HRBF) model and it grows by automatically adding units at smaller scales, where the surface details are located, while data points are being collected. Real-time operation is achieved by exploiting the quasi-local nature of the Gaussian units: through the definition of a quad-tree structure to support their receptive field local network reconfiguration can be obtained. The model has been applied to 3-D scanning, where an updated real-time display of the manifold to the operator is fundamental to drive the acquisition procedure itself. Quantitative results are reported, which show that the accuracy achieved is comparable to that of two batch approaches: batch HRBF and support vector machines (SVMs). However, these two approaches are not suitable to real-time online learning. Moreover, proof of convergence is also given.      

崎岖地表上的旋翼无人机自主安全降落系统

针对旋翼无人机全自主作业的需求,构建了崎岖地表上的旋翼无人机自主安全降落系统．该系统通过机载实时运算自动分析落区地形,寻找可行落点并实施自动降落．系统以低成本的立体RGB-D相机作为深度传感设备,利用截断符号距离函数（TSDF）对着陆区地形进行实时3维建模,生成低噪的落区地形深度图像,并设计了一种适应起落机构形状的实时精细落点搜索方法,最后使用级联PID（比例－积分－微分）控制器控制无人机实施安全降落．系统基于大疆M100无人机平台实现,定制了仿真器进行算法调试,并最终在实际的崎岖地表上实现了自主安全降落．本文工作可为旋翼无人机紧急降落、物流运输或者灾后搜救提供有效安全的解决方案．