Shape recovery from turntable sequence using rim reconstruction

This paper makes use of both feature points and silhouettes to deliver fast 3D shape recovery. The algorithm exploits object silhouettes in two views to establish a 3D rim curve, which is defined with respect to the two frontier points arising from two views. The images of this 3D rim curve in the two views are matched using cross-correlation technique. A 3D planar rim curve is then reconstructed using point-based reconstruction method. A set of 3D rim curves enclosing the object can be obtained from an image sequence captured under circular motion. Silhouettes are further utilized to check for mismatched rim points. The proposed method solves the problem of reconstruction of concave object surface, which is usually left unresolved in general silhouette-based reconstruction methods. In addition, the property of the reconstructed 3D rim curves allows fast surface extraction. Experimental results with real data are presented.     

面向点云的三维物体识别方法综述

随着三维扫描技术的快速发展,获取各类场景的点云数据已经非常简单快捷;加之点云数据具备不受光照、阴影、纹理的影响等优势,基于点云的三维物体识别已成为计算机视觉领域的研究热点。首先,对近年来面向点云数据的三维物体识别方法进行归纳和总结;然后,对已有方法的优势及缺点进行分析;最后,指出点云物体识别中所面临的挑战及进一步的研究方向。     

A review of 3D reconstruction techniques in civil engineering and their applications

Three-dimensional (3D) reconstruction techniques have been used to obtain the 3D representations of objects in civil engineering in the form of point cloud models, mesh models and geometric models more often than ever, among which, point cloud models are the basis. In order to clarify the status quo of the research and application of the techniques in civil engineering, literature retrieval is implemented by using major literature databases in the world and the result is summarized by analyzing the abstracts or the full papers when required. First, the research methodology is introduced, and the framework of 3D reconstruction techniques is established. Second, 3D reconstruction techniques for generating point clouds and processing point clouds along with the corresponding algorithms and methods are reviewed respectively. Third, their applications in reconstructing and managing construction sites and reconstructing pipelines of Mechanical, Electrical and Plumbing (MEP) systems, are presented as typical examples, and the achievements are highlighted. Finally, the challenges are discussed and the key research directions to be addressed in the future are proposed. This paper contributes to the knowledge body of 3D reconstruction in two aspects, i.e. summarizing systematically the up-to-date achievements and challenges for the applications of 3D reconstruction techniques in civil engineering, and proposing key future research directions to be addressed in the field.     

Point cloud simplification for the boundary preservation based on extracted four features

This paper proposes a simplification algorithm based on four feature parameters, aiming at solving the problem that the edge features cannot be retained due to the incompletely extracted sharp features during point cloud simplification. Firstly, K neighborhood searching is carried out for point cloud, and K neighborhood points are quickly found by a dynamic grid method. Then, four features including: the curvature of the point, the average of the normal angle of a point from a neighborhood point, the average distance between the point and the neighborhood point and the distance between the point and the center of gravity of the neighborhood point, are calculated according to the K neighborhood of the data point. The four parameters are used to define the feature discrimination parameters and feature thresholds, to compare the size and extract the feature points; finally, the non-feature points are reduced twice by the method of the bounding box, and the reduced point cloud and feature points are spliced to achieve the purpose of simplification. The experimental results show that the distance between the point and the center of gravity of the neighborhood has a great influence on the simplified model boundary, which effectively guarantees the accuracy of the simplified model.     

Video summarization via minimum sparse reconstruction

The rapid growth of video data demands both effective and efficient video summarization methods so that users are empowered to quickly browse and comprehend a large amount of video content. In this paper, we formulate the video summarization task with a novel minimum sparse reconstruction (MSR) problem. That is, the original video sequence can be best reconstructed with as few selected keyframes as possible. Different from the recently proposed convex relaxation based sparse dictionary selection method, our proposed method utilizes the true sparse constraint L₀ norm, instead of the relaxed constraint _L2,1$L_{2,1}$ norm, such that keyframes are directly selected as a sparse dictionary that can well reconstruct all the video frames. An on-line version is further developed owing to the real-time efficiency of the proposed MSR principle. In addition, a percentage of reconstruction (POR) criterion is proposed to intuitively guide users in obtaining a summary with an appropriate length. Experimental results on two benchmark datasets with various types of videos demonstrate that the proposed methods outperform the state of the art.     

Image segmentation using a local extrema texture measure

An image segmentation technique is proposed which uses a texture measure that counts the number of local extrema in a window centered at each picture point. Four gray level pictures are derived, each of which represents a texture or gray level property of the original image. These intermediate pictures are used to derive the number of segments in which to divide the original image. The segmentation is then performed by assigning each pixel in the original image to a region by using a four-dimensional distance measure on the intermediate pictures, comparing each location to each selected segment. This process is then repeated in a hierarchical structure using decreasing window sizes so that smaller regions within the larger ones are defined. The computations required are amenable to real-time video implementation using state-of-the-art devices.     

Salient object detection via local saliency estimation and global homogeneity refinement

This paper presents a new hybrid approach for detecting salient objects in an image. It consists of two processes: local saliency estimation and global-homogeneity refinement. We model the salient object detection problem as a region growing and competition process by propagating the influence of foreground and background seed-patches. First, the initial local saliency of each image patch is measured by fusing local contrasts with spatial priors, thereby the seed-patches of foreground and background are constructed. Later, the global-homogeneous information is utilized to refine the saliency results by evaluating the ratio of the foreground and background likelihoods propagated from the seed-patches. Despite the idea is simple, our method can effectively achieve consistent performance for detecting object saliency. The experimental results demonstrate that our proposed method can accomplish remarkable precision and recall rates with good computational efficiency.     

融合目标增强与稀疏重构的显著性检测

目的 为了解决图像显著性检测中存在的边界模糊,检测准确度不够的问题,提出一种基于目标增强引导和稀疏重构的显著检测算法（OESR）。方法 基于超像素,首先从前景角度计算超像素的中心加权颜色空间分布图,作为前景显著图;由图像边界的超像素构建背景模板并对模板进行预处理,以优化后的背景模板作为稀疏表示的字典,计算稀疏重构误差,并利用误差传播方式进行重构误差的校正,得到背景差异图;最后,利用快速目标检测方法获取一定数量的建议窗口,由窗口的对象性得分计算目标增强系数,以此来引导两种显著图的融合,得到最终显著检测结果。结果 实验在公开数据集上与其他12种流行算法进行比较,所提算法对具有不同背景复杂度的图像能够较准确的检测出显著区域,对显著对象的提取也较为完整,并且在评价指标检测上与其他算法相比,在MSRA10k数据集上平均召回率提高4.1%,在VOC2007数据集上,平均召回率和F检验分别提高18.5%和3.1%。结论 本文提出一种新的显著检测方法,分别利用颜色分布与对比度方法构建显著图,并且在显著图融合时采用一种目标增强系数,提高了显著图的准确性。实验结果表明,本文算法能够检测出更符合视觉特性的显著区域,显著区域更加准确,适用于自然图像的显著性目标检测、目标分割或基于显著性分析的图像标注。     

Image segmentation by pixel classification

A fast boundary finding algorithm is presented which works without threshold operation and without any interactive control. The procedure can be described as a hierarchical two-step algorithm. In the first step the image is divided into two disjunct regions, one of them including the whole object of interest.In the second step the problem of boundary finding is suggested as a classification problem, which means that for any pixel a four-dimensional feature vector is computed which allows classification of pixels into contour elements and any other pixels.The algorithm was tested on several thousand cell images and can be easily adapted to other problems by modification of a set of parameters.     

Advances in 3D data acquisition and processing for industrial applications

A critical task of vision-based manufacturing applications is to generate a virtual representation of a physical object from a dataset of point clouds. Its success relies on reliable algorithms and tools. Many effective technologies have been developed to solve various problems involved in data acquisition and processing. Some articles are available on evaluating and reviewing these technologies and underlying methodologies. However, for most practitioners who lack a strong background on mathematics and computer science, it is hard to understand theoretical fundamentals of the methodologies. In this paper, we intend to survey and evaluate recent advances in data acquisition and progressing, and provide an overview from a manufacturing perspective. Some potential manufacturing applications have been introduced, the technical gaps between the practical requirements and existing technologies discussed, and research opportunities identified.     

A robot hand-eye calibration method of line laser sensor based on 3D reconstruction

In the robotic eye-in-hand measurement system, a hand-eye calibration method is essential. From the perspective of 3D reconstruction, this paper first analyzes the influence of the line laser sensor hand-eye calibration error on the 3D reconstructed point clouds error. Based on this, considering the influence of line laser sensor measurement errors and the need for high efficiency and convenience in robotic manufacturing systems, this paper proposes a 3D reconstruction-based robot line laser hand-eye calibration method. In this method, combined with the point cloud registration technique, the newly defined error-index more intuitively reflects the calibration result than traditional methods. To raise the performance of the calibration algorithm, a Particle Swarm Optimization - Gaussian Process (PSO-GP) method is adopted to improve the efficiency of the calibration. The experiments show that the Root Mean Square Error (RMSE) of the reconstructed point cloud can reach 0.1256 mm when using the proposed method, and the reprojection error is superior to those using traditional hand-eye calibration methods.     

Point Cloud Denoising via Moving RPCA

We present an algorithm for the restoration of noisy point cloud data, termed Moving Robust Principal Components Analysis (MRPCA). We model the point cloud as a collection of overlapping two‐dimensional subspaces, and propose a model that encourages collaboration between overlapping neighbourhoods. Similar to state‐of‐the‐art sparse modelling‐based image denoising, the estimated point positions are computed by local averaging. In addition, the proposed approach models grossly corrupted observations explicitly, does not require oriented normals, and takes into account both local and global structure. Sharp features are preserved via a weighted ?₁ minimization, where the weights measure the similarity between normal vectors in a local neighbourhood. The proposed algorithm is compared against existing point cloud denoising methods, obtaining competitive results.     

Logarithmic dyadic wavelet transform with its applications in edge detection and reconstruction

In this paper, based on the logarithmic image processing model and the dyadic wavelet transform (DWT), we introduce a logarithmic DWT (LDWT) that is a mathematical transform. It can be used in image edge detection, signal and image reconstruction. Comparative study of this proposed LDWT-based method is done with the edge detection Canny and Sobel methods using Pratt's Figure of Merit, and the comparative results show that the LDWT-based method is better and more robust in detecting low contrast edges than the other two methods. The gradient maps of images are detected by using the DWT- and LDWT-based methods, and the experimental results demonstrate that the gradient maps obtained by the LDWT-based method are more adequate and precisely located. Finally, we use the DWT- and LDWT-based methods to reconstruct one-dimensional signals and two-dimensional images, and the reconstruction results show that the LDWT-based reconstruction method is more effective.     

新的连通域标记方法及其在医学图像中的应用*

为了提高诊断的准确率和效率,提出了一种新的连通域标记算法,同时对医学图像中感兴趣区域进行连通域标记和区域特征提取。该算法先对读入二值图进行边界提取,再对边界进行跟踪和标记,利用图像重构的方法对边界进行区域填充,并将属于同一连通域的边界进行归类,即重新排列标记号,最后对连通区域的形态特征进行提取。实验证明,该算法不但能正确标记任意复杂形状的连通域,运行速度较快,而且对连通区域进行了特征提取,现已应用到医学图像处理的多个方面,为下一步的图像处理奠定了更好的基础。