图卷积网络下牙齿种子点自动选取

目的 选取牙齿种子点是计算机正畸中常用牙齿分割方法的关键步骤。目前业内大部分牙齿正畸软件都采用需要交互标记的分割方法,通过人机交互在3维牙颌模型上选取每一颗牙齿的种子点,效率较低。针对这一问题,提出基于特征导向的图卷积网络（feature-steered graph convolutional network,FeaStNet）牙齿种子点自动选取方法。方法 通过分析每个牙齿类型的种子点位置和最终分割效果,设立统一的规则,建立了牙颌模型的种子点数据集;利用特征导向的图卷积构建了新的多尺度网络结构,用于识别3维牙颌模型上的特征信息,为了更好地拟合牙齿特征,加深网络模型的深度;再通过训练调整参数和多尺度网络结构,寻找特定的种子点,使用均值平方差损失函数对模型进行评估,以提高预测模型的精确度;把网络寻找出的特征点作为基础点,在牙颌模型上找出与基础点距离最近的点作为种子点,如果种子点位置准确,则根据种子点将牙齿与牙龈分割开。对于种子点位置不准确的结果,通过人工操作修正种子点位置,再进行分割。结果 实验在自建的数据集中测试,其中种子点全部准确的牙颌占88%,其余情况下只需要调整部分不准确种子点的位置。该方法简单快速,与现有方法相比,需要较少的人工干预,提高了工作效率。结论 提出的种子点自动选取方法,能够自动选取牙齿种子点,解决牙齿分割中需要进行交互标记的问题,基本实现了牙齿分割的自动化,适用于各类畸形牙患者模型的牙齿分割。     

从机载激光扫描数据中交互式提取人字形房屋模型的方法研究

快速有效地从机载激光扫描(airborne lidar)点云数据中提取房屋模型是机载激光扫描系统应用研究的一项重要课题。鉴于交互式半自动方法是从点云数据中提取简单规则房屋模型信息的一种可行的方法,为此采用3维空间中改进的Hough变换以及聚类分析,提出了一种从点云数据中交互式提取人字形房屋模型的方法。该方法分为3个步骤:第1步是用户确定房屋区域,并分割出候选的屋顶点集;第2步是对候选屋顶点集采用3维空间中改进的Hough变换,然后对Hough变换后所获得的参数集进行聚类分析,以此获得屋顶所在平面的参数表达;第3步是构造完整的房屋模型。通过屋顶平面相交得到屋脊线,通过点的范围分析确定屋顶的边缘,最后添加竖直的墙面构造完整的房屋模型。经采用Optech公司提供的数据进行实验初步证实,该方法是可行的,且整个提取过程只需要很少的用户交互,因此适合于大规模处理机载激光扫描数据。     

基于显著性与脉冲耦合神经网络的图像分割

针对复杂图像易受背景干扰的问题,提出一种基于显著性与脉冲耦合神经网络(Saliency and Pulse Coupled Neural Network,SPCNN)的图像分割方法。首先,利用显著性检测算法和最大类间方差法获得显著性图以及目标图像,排除了背景对初始种子点选取的干扰;然后,计算出显著性图的质心,并将其作为初始种子点;最后,采用改进的基于区域生长的脉冲耦合神经网络对目标图像进行分割。在Berkeley图像库和Ground truth Database图像库上对SPCNN模型进行了验证。实验结果表明,在一致性系数CC、相似性系数SC、综合指标IC 3个方面,SPCNN模型均优于所对比的PCNN模型、区域生长模型和RG-PCNN模型。     

多颜色空间上的交互式图像分割

介绍了视觉颜色空间及其在交互式图像分割中的作用,实验分析了它的奇异性,在此基础上,考虑像素的空间和色彩分布,提出了基于区域生长法的多颜色空间、多度量准则的聚类算法和零碎区域的合并算法,颜色空间选取HSL和RGB两种,相似性度量包括了种子点、扩张点和生长区域三个方面,并用于敦煌壁画图像的分割。     

Robust normal estimation and region growing segmentation of infrastructure 3D point cloud models

Modern remote sensing technologies such as three-dimensional (3D) laser scanners and image-based 3D scene reconstruction are in increasing demand for applications in civil infrastructure design, maintenance, operation, and as-built construction verification. The complex nature of the 3D point clouds these technologies generate, as well as the often massive scale of the 3D data, make it inefficient and time consuming to manually analyze and manipulate point clouds, and highlights the need for automated analysis techniques. This paper presents one such technique, a new region growing algorithm for the automated segmentation of both planar and non-planar surfaces in point clouds. A core component of the algorithm is a new point normal estimation method, an essential task for many point cloud processing algorithms. The newly developed estimation method utilizes robust multivariate statistical outlier analysis for reliable normal estimation in complex 3D models, considering that these models often contain regions of varying surface roughness, a mixture of high curvature and low curvature regions, and sharp features. An adaptation of Mahalanobis distance, in which the mean vector and covariance matrix are derived from a high-breakdown multivariate location and scale estimator called Deterministic MM-estimator (DetMM) is used to find and discard outlier points prior to estimating the best local tangent plane around any point in a cloud. This approach is capable of more accurately estimating point normals located in highly curved regions or near sharp features. Thereafter, the estimated point normals serve a region growing segmentation algorithm that only requires a single input parameter, an improvement over existing methods which typically require two control parameters. The reliability and robustness of the normal estimation subroutine was compared against well-known normal estimation methods including the Minimum Volume Ellipsoid (MVE) and Minimum Covariance Determinant (MCD) estimators, along with Maximum Likelihood Sample Consensus (MLESAC). The overall region growing segmentation algorithm was then experimentally validated on several challenging 3D point clouds of real-world infrastructure systems. The results indicate that the developed approach performs more accurately and robustly in comparison with conventional region growing methods, particularly in the presence of sharp features, outliers and noise.     

Area-wide roof plane segmentation in airborne LiDAR point clouds

Most algorithms performing segmentation of 3D point cloud data acquired by, e.g. Airborne Laser Scanning (ALS) systems are not suitable for large study areas because the huge amount of point cloud data cannot be processed in the computer’s main memory. In this study a new workflow for seamless automated roof plane detection from ALS data is presented and applied to a large study area. The design of the workflow allows area-wide segmentation of roof planes on common computer hardware but leaves the option open to be combined with distributed computing (e.g. cluster and grid environments). The workflow that is fully implemented in a Geographical Information System (GIS) uses the geometrical information of the 3D point cloud and involves four major steps: (i) The whole dataset is divided into several overlapping subareas, i.e. tiles. (ii) A raster based candidate region detection algorithm is performed for each tile that identifies potential areas containing buildings. (iii) The resulting building candidate regions of all tiles are merged and those areas overlapping one another from adjacent tiles are united to a single building area. (iv) Finally, three dimensional roof planes are extracted from the building candidate regions and each region is treated separately. The presented workflow reduces the data volume of the point cloud that has to be analyzed significantly and leads to the main advantage that seamless area-wide point cloud based segmentation can be performed without requiring a computationally intensive algorithm detecting and combining segments being part of several subareas (i.e. processing tiles). A reduction of 85% of the input data volume for point cloud segmentation in the presented study area could be achieved, which directly decreases computation time.     

基于最大内切圆的肝影像自动分割方法

针对肝脏区域提取中的手动选择种子点以及提取时的准确性和完整性问题，提出了一种基于最大内切圆的肝影像自动分割算法。采用最大区域面积测量法锁定二值化后的肝脏区域，通过寻找锁定肝脏区域的最大内切圆的圆心来自动获取种子点位置；采用改进的区域生长算法进行图像分割。实验结果表明，该方法有效地解决了区域生长的种子点手动选取问题，并且能够精确、完整地分割出肝脏组织，避免了受主观因素影响而将种子点选取在边缘或噪声等错误位置。     

Three-dimensional point cloud plane segmentation in both structured and unstructured environments

This paper focuses on three-dimensional (3D) point cloud plane segmentation. Two complementary strategies are proposed for different environments, i.e., a subwindow-based region growing (SBRG) algorithm for structured environments, and a hybrid region growing (HRG) algorithm for unstructured environments. The point cloud is decomposed into subwindows first, using the points’ neighborhood information when they are scanned by the laser range finder (LRF). Then, the subwindows are classified as planar or nonplanar based on their shape. Afterwards, only planar subwindows are employed in the former algorithm, whereas both kinds of subwindows are used in the latter. In the growing phase, planar subwindows are investigated directly (in both algorithms), while each point in nonplanar subwindows is investigated separately (only in HRG). During region growing, plane parameters are computed incrementally when a subwindow or a point is added to the growing region. This incremental methodology makes the plane segmentation fast. The algorithms have been evaluated using real-world datasets from both structured and unstructured environments. Furthermore, they have been benchmarked against a state-of-the-art point-based region growing (PBRG) algorithm with regard to segmentation speed. According to the results, SBRG is 4 and 9 times faster than PBRG when the subwindow size is set to 3×3 and 4×4 respectively; HRG is 4 times faster than PBRG when the subwindow size is set to 4×4. Open-source code for this paper is available at https://github.com/junhaoxiao/TAMS-Planar-Surface-Based-Perception.git.     

Pictorial recognition of objects employing affine invariance in thefrequency domain

Describes an efficient approach to pose invariant pictorial object recognition employing spectral signatures of image patches that correspond to object surfaces which are roughly planar. Based on singular value decomposition (SVD), the affine transform is decomposed into slant, tilt, swing, scale, and 2D translation. Unlike previous log-polar representations which were not invariant to slant, our log-log sampling configuration in the frequency domain yields complete affine invariance. The images are preprocessed by a novel model-based segmentation scheme that detects and segments objects that are affine-similar to members of a model set of basic geometric shapes. The segmented objects are then recognized by their signatures using multidimensional indexing in a pictorial dataset represented in the frequency domain. Experimental results with a dataset of 26 models show 100 percent recognition rates in a wide range of 3D pose parameters and imaging degradations: 0-360° swing and tilt, 0-82° of slant, more than three octaves in scale change, window-limited translation, high noise levels (0 dB), and significantly reduced resolution (1:5)     

Urban building roof segmentation from airborne lidar point clouds

This article presents a new approach to segmenting building rooftops from airborne lidar point clouds. A progressive morphological filter technique is first applied for separation between ground and non-ground points. For the non-ground points, a region-growing algorithm based on a plane-fitting technique is used to separate building points from vegetation points. Then, an adaptive Random Sample Consensus (RANSAC) algorithm based on a grid structure is developed to improve the probability of selecting an uncontained sample from the localized sampling. The distance, standard deviation and normal vector are integrated to keep topological consistency among building rooftop patches during building rooftop segmentation. Finally, the remaining points are mapped on to the extracted planes by a post-processing technique to improve the segmentation accuracy. The results for buildings with different roof complexities are presented and evaluated.     

建筑物的着色点云平面区域分割研究

从激光点云中提取建筑物平面区域是当前建筑物三维建模的关键技术。通过分析激光点云数据中建筑物的特征,引入k-d tree数据结构及随机霍夫变换（RHT）算法,提出了融合几何与颜色相似信息的区域生长分割算法。为了避免分割过程中的过度分割和分割不足,算法需要人工设置几个参数。通过在一组建筑物中提取实验数据,验证了该算法的有效性。     

Complete classification of raw LIDAR data and 3D reconstruction of buildings

LIDAR (LIght Detection And Ranging) data are a primary data source for digital terrain model (DTM) generation and 3D city models. This paper presents a three-stage framework for a robust automatic classification of raw LIDAR data as buildings, ground and vegetation, followed by a reconstruction of 3D models of the buildings. In the first stage the raw data are filtered and interpolated over a grid. In the second stage, first a double raw data segmentation is performed and then geometric and topological relationships among regions resulting from segmentation are computed and stored in a knowledge base. In the third stage, a rule-based scheme is applied for the classification of the regions. Finally, polyhedral building models are reconstructed by analysing the topology of building outlines, building roof slopes and eaves lines. Results obtained on data sets with different ground point density, gathered over the town of Pavia (Italy) with Toposys and Optech airborne laser scanning systems, are shown to illustrate the effectiveness of the proposed approach.     

Multi-object reconstruction from dynamic scenes: An object-centered approach

In this paper, we present a new framework for three-dimensional (3D) reconstruction of multiple rigid objects from dynamic scenes. Conventional 3D reconstruction from multiple views is applicable to static scenes, in which the configuration of objects is fixed while the images are taken. In our framework, we aim to reconstruct the 3D models of multiple objects in a more general setting where the configuration of the objects varies among views. We solve this problem by object-centered decomposition of the dynamic scenes using unsupervised co-recognition approach. Unlike conventional motion segmentation algorithms that require small motion assumption between consecutive views, co-recognition method provides reliable accurate correspondences of a same object among unordered and wide-baseline views. In order to segment each object region, we benefit from the 3D sparse points obtained from the structure-from-motion. These points are reliable and serve as automatic seed points for a seeded-segmentation algorithm. Experiments on various real challenging image sequences demonstrate the effectiveness of our approach, especially in the presence of abrupt independent motions of objects.     

A simple method of rapid and automatic color image segmentation for serialized Visible Human slices

In this paper, a rapid and automatic color image segmentation method for the serialized slices of the Visible Human is proposed. The main strategy is based on region growing and pixel color difference. A rapid color similarity computing method is improved and applied for classifying different pixels. An algorithm based on corrosion from four directions is proposed to automatically extract the seed points for the serialized slices. Utilizing this method, the color slice images of the Visible Human body can be segmented in series automatically. Also, the multithreading frame of parallel computing is introduced in the entire segmentation process. This method is simple but rapid and automatic. The primary organs of the Visible Human can be segmented clearly and accurately. The 3D models of these organs after 3D reconstruction are satisfactory. This novel method can provide support to the Visible Human research.     

Automatic generation of high-quality building models from lidar data

Automating data acquisition for 3D city models is an important research topic in photogrammetry. In addition to techniques that rely on aerial images, generating 3D building models from point clouds provided. by light detection and ranging (Lidar) sensors is gaining importance. The progress in sensor technology has triggered this development. Airborne laser scanners can deliver dense point clouds with densities of up to one point per square meter. Using this information, it's possible to detect buildings and their approximate outlines and also to extract planar roof faces and create models that correctly resemble the roof structures. The author presents a method for automatically generating 3D building models from point clouds generated by the Lidar sensing technology.     

Automatic detection of building roofs from point clouds produced by the dense image matching technique

Three-dimensional (3D) spatial information of object points is a vital requirement for many disciplines. Laser scanning technology and techniques based on image matching have been used extensively to produce 3D dense point clouds. These data are used frequently in various applications, such as the generation of digital surface model (DSM)/digital terrain model (DTM), extracting objects (e.g., buildings, trees, and roads), 3D modelling, and detecting changes. The aim of this study was to extract the building roof points automatically from the 3D point cloud data created via the image matching techniques with optical aerial images (with red, green, and blue band (RGB) and infrared (IR)). In the first stage of the study, as an alternative to laser scanning technology, which is more expensive than optical imaging systems, the 3D point clouds were produced by matching high-resolution images using a Semi Global Matching algorithm. The normalized difference vegetation index (NDVI) values for each point were calculated using the spectral information (RGB + IR) in the 3D point cloud data, and the points that represented the vegetation cover were determined using these values. In the second stage, existing ground and non-ground points that were free of vegetation cover were determined within the point cloud. Subsequently, only the points on the roof of the building were detected automatically using the proposed algorithm. Thus, points of the roofs of buildings located in areas with different topographic characteristics were detected automatically detected using only images. It was determined that the average values of correctness (Corr), completeness (Comp), and quality (Q) of the pixel-based accuracy analysis metrics were 95%, 98%, and 93%, respectively, in the selected test areas. According to the results of the accuracy analysis, it is clear that the proposed algorithm is very successful in automatic extraction of building roof points.     

一种变形表面的三维测量方法

提出一种新的利用标定图像进行三维测量的方法。利用SIFT算法找到初始的对应点,然后根据这些点生成三维空间中的种子点,再以这些种子点为中心,向外区域增长,直到完成整个物体表面测量。在每次增长的过程中,需要计算增长的三维空间平面在两个相机上的投影之间的图像相关系数。图像相关系数较大时认为是正确的增长,否则是错误的增长。实验证明,使用该方法能够得到很好的三维测量结果。     

伪点云修正增强激光雷达数据

目的 激光雷达在自动驾驶中具有重要意义,但其价格昂贵,且产生的激光线束数量仍然较少,造成采集的点云密度较稀疏。为了更好地感知周围环境,本文提出一种激光雷达数据增强算法,由双目图像生成伪点云并对伪点云进行坐标修正,进而实现激光雷达点云的稠密化处理,提高3D目标检测精度。此算法不针对特定的3D目标检测网络结构,是一种通用的点云稠密化方法。方法 首先利用双目RGB图像生成深度图像,根据先验的相机参数和深度信息计算出每个像素点在雷达坐标系下的粗略3维坐标,即伪点云。为了更好地分割地面,本文提出了循环RANSAC （random sample consensus）算法,引入了一个分离平面型非地面点云的暂存器,改进复杂场景下的地面分割效果。然后将原始点云进行地面分割后插入KDTree （k-dimensional tree）,以伪点云中的每个点为中心在KDTree中搜索若干近邻点,基于这些近邻点进行曲面重建。根据曲面重建结果,设计一种计算几何方法导出伪点云修正后的精确坐标。最后,将修正后的伪点云与原始激光雷达点云融合得到稠密化点云。结果 实验结果表明,稠密化的点云在视觉上具有较好的质量,物体具有更加完整的形状和轮廓,并且在KITTI （Karlsruhe Institute of Technology and Toyota Technological Institute）数据集上提升了3D目标检测精度。在使用该数据增强方法后,KITTI数据集下AVOD （aggregate view object detection）检测方法的AP_3D-Easy （average precision of 3D object detection on easy setting）提升了8.25%,AVOD-FPN （aggregate view object detection with feature pyramid network）检测方法的AP_BEV-Hard （average precision of bird’s eye view on hard setting）提升了7.14%。结论 本文提出的激光雷达数据增强算法,实现了点云的稠密化处理,并使3D目标检测结果更加精确。