基于改进Census变换的多特性立体匹配算法

针对当前立体匹配算法存在的匹配准确率低，难以达到实用的高精度水平的问题，提出了一种基于改良的Census变换与色彩信息和梯度测度相结合的多特性立体匹配算法，实现高精度的双目立体匹配。算法首先在初始代价匹配阶段，将改进的Census变换、色彩和梯度测度赋权求和得出可靠的初始匹配代价；在聚合阶段，采取高效快捷的最小生成树聚合，获得匹配代价矩阵；最后根据胜者为王法则得到初始视差图，并引入左右一致性检测等策略优化视差图，获得高精度的视差图，实验阶段对源自Middlebury上的标准测试图进行测试验证，实验结果表明，经本文算法处理得到的15组测试数据集的视差图在非遮挡区域的平均误匹配率为6.81%，算法实时响应性优良。     

一种改进的实时半全局立体匹配算法及硬件实现

在基于现场可编程门阵列的实时立体匹配系统中,Census变换算法针对特定区域的误匹配率较高。为提高匹配精度,提出一种具有高并行性流水线结构的实时半全局立体匹配算法并进行硬件实现。将改进的Tanimoto距离和带权重4方向的梯度绝对值差进行组合,作为新的初始匹配代价。在代价聚合阶段采用4路径并行结构的SGM算法,在视差选择阶段采用赢家通吃策略,在视差校正阶段采用阈值检测算法代替传统左右一致性检验算法。实验结果表明,该算法能够有效提高弱纹理和边缘区域的区分度,减少对中心点的依赖,降低资源占用,其在Middleburry平台上的平均误匹配率仅为7.52%,在Xilinx Zynq-7000平台上的匹配速率达到98 frame/s。     

基于mean-shift全局立体匹配方法

针对图像全局立体匹配精度高、计算量大的问题,提出基于mean shift图像分割的全局立体匹配方法。首先,通过mean shift算法对图像进行分割,获取图像同质区域数量和区域的标号。在计算匹配代价时,根据像素所属的分割区域,对像素进行筛选,从而提高匹配代价计算速度;其次,在代价聚合前,将mean shift算法获取的同质区域数K值赋值给K-means聚类算法,对像素再次聚类,提高立体匹配精度和速度;最后通过TRW-S置信传播解决能量最小化问题。实验表明,该算法明显提高了匹配的准确性和速度,与单纯的全局匹配算法相比,具有更大的优势。     

融合梯度信息自适应窗口的立体匹配

针对立体匹配算法中对固定窗口匹配代价计算存在匹配精度低的问题，该文提出了一种基于融合梯度信息自适应窗口的立体匹配算法。以AD-Census和SGM立体匹配算法为基础，首先研究了窗口大小对算法性能的影响，并在代价计算中加入融合梯度信息的自适应窗口算法，使用标准图像集对改进前后的算法性能进行对比。实验结果表明，匹配窗口增大可以提高算法的匹配精度但会降低算法效率，利用梯度信息调节窗口大小可以提高算法的适用性，保持效率的同时降低三类区域中的误匹配率。在4个测试集中误匹配率平均下降了4.05%，平均运行时间只增加了0.82 s，改进后的立体匹配算法具有更好的综合性能。     

融合多特征表示和超像素优化的双目立体匹配

针对目前许多局部双目立体匹配方法在缺乏纹理区域、遮挡区域、深度不连续区域匹配精度低的问题,提出了基于多特征表示和超像素优化的立体匹配算法。通过在代价计算步骤中加入边缘信息特征,与图像局部信息代价相融合,增加了在视差计算时边缘区域的辨识度;在代价聚合步骤,基于超像素分割形成的超像素区域,利用米字骨架自适应搜索,得到聚合区域,对初始代价进行聚合;在视差精化步骤利用超像素分割信息,对匹配错误视差进行修正,提高匹配精度。基于Middlebury立体视觉数据集测试平台,与自适应权重AD-Census、FA等方法得出的视差图进行比较,该算法在深度不连续区域和缺乏纹理区域的匹配效果显著改善,提高了立体匹配精度。     

基于融合代价和优化引导滤波的立体匹配算法

现有局部立体匹配算法与全局立体匹配算法相比,计算量更小、速度更快,能达到实时匹配的要求,但存在误匹配率较高、视差结果精度较低等问题。提出一种新的局部立体匹配算法,依据像素梯度信息、像素梯度的平均值及标准差计算多指标梯度代价,使算法对图像局部结构的表达更加全面。根据像素的灰度差异程度划分等级,提出7等级编码的精细化Census变换匹配代价,以有效识别图像信息之间的细微差异,弥补Census变换在相似纹理区域精度较低的不足。将多指标梯度代价和精细化Census变换匹配代价按一定权重进行融合,形成初始匹配代价,从而充分发挥两种代价的优越性。在代价聚合阶段,对引导滤波的线性系数进行自适应优化,解决固定参数引起图像局部过于平滑或平滑不足的问题,并使用优化后的引导滤波模型进行代价聚合,改善代价聚合的效果。使用胜者为王算法计算视差获得初始视差图,最后通过左右一致性检测和加权中值滤波进行视差优化,得到更为理想的视差图。在Middlebury V3立体评估平台上的实验结果表明,所提算法在非遮挡区域的加权平均误匹配率为15.7%,与Cens5、IGF、ISM等算法相比具有较高的精度。     

多元线性回归引导的立体匹配算法

深度图像中视差跳变的像素点匹配一直是立体匹配的挑战性问题之一.基于引导滤波的局部立体匹配算法通过考虑匹配图像内容,可以在保持深度图像边缘的同时提高匹配精度、加快匹配速度,但引导滤波会产生图像光晕,在图像边缘区域也会引入大量的噪声.为此,将引导滤波的岭回归扩展到多元回归,提出一种基于多元线性回归的立体匹配算法.首先将引导滤波中只含图像像素值这一单变量的回归方程扩展为基于图像像素值和梯度信息等多个变量的多元回归方程,对初始代价值进行滤波聚合,并与单独进行引导滤波的匹配代价聚合值进行加权组合提高图像边缘的匹配效果;然后根据代价聚合最小值与次小值之间的相互关系定义了视差选择可信度,解决了视差选择时的歧义问题.在Middlebury测试平台进行了实验的结果表明,文中算法有效地提高深度图像中视差跳变像素点的匹配精度,降低了匹配噪声;与最新的高性能立体匹配算法相比,该算法可以以较小的计算复杂度获得高质量的视差图.     

多种相似性测度结合的局部立体匹配算法

针对立体匹配中匹配代价和支持窗口难以选择的问题,提出一种将多种相似性测度相结合的局部立体匹配算法.首先,构造匹配代价,结合图像的Census变换、WLD(Weber局部描述符)特征、图像色彩信息以及图像梯度信息作为匹配代价;然后,使用引导滤波器对匹配代价进行聚合;最后,针对WTA(赢者全取)策略引入的视差选择歧义问题和左右一致性检测(LRC)引入的水平条纹问题,提出了一种基于可信度和加权滤波的视差修正算法.利用Middlebury测试平台提供的标准测试图像对本文算法进行测试,其平均错误匹配率为5.30%,与Fast Bilateral算法等一些公认的性能优异算法相比,本文算法提高了匹配准确率.     

融合边缘保持与改进代价聚合的立体匹配算法

目的 立体匹配是计算机双目视觉的重要研究方向,主要分为全局匹配算法与局部匹配算法两类。传统的局部立体匹配算法计算复杂度低,可以满足实时性的需要,但是未能充分利用图像的边缘纹理信息,因此在非遮挡、视差不连续区域的匹配精度欠佳。为此,提出了融合边缘保持与改进代价聚合的立体匹配。方法 首先利用图像的边缘空间信息构建权重矩阵,与灰度差绝对值和梯度代价进行加权融合,形成新的代价计算方式,同时将边缘区域像素点的权重信息与引导滤波的正则化项相结合,并在多分辨率尺度的框架下进行代价聚合。所得结果经过视差计算,得到初始视差图,再通过左右一致性检测、加权中值滤波等视差优化步骤获得最终的视差图。结果 在Middlebury立体匹配平台上进行实验,结果表明,融合边缘权重信息对边缘处像素点的代价量进行了更加有效地区分,能够提升算法在各区域的匹配精度。其中,未加入视差优化步骤的21组扩展图像对的平均误匹配率较改进前减少3.48%,峰值信噪比提升3.57 dB,在标准4幅图中venus上经过视差优化后非遮挡区域的误匹配率仅为0.18%。结论 融合边缘保持的多尺度立体匹配算法有效提升了图像在边缘纹理处的匹配精度,进一步降低了非遮挡区域与视差不连续区域的误匹配率。     

基于自适应权重的立体匹配优化算法

为解决现有立体匹配算法对低纹理以及视差不连续区域匹配效果较差的问题,提出一种改进的立体匹配优化算法。在传统自适应权重算法匹配代价的基础上,融合高斯差分图像差分信息,即左右图像高斯差分图的差分,重新定义其初始匹配代价,增加算法在视差不连续区域的鲁棒性,并加入边缘约束和视差边缘约束迭代聚类以及基于高斯差分图的自适应窗口算法,保证改进算法在低纹理区域的匹配性能,消除坏点与视差空洞。将该算法与传统自适应权重匹配算法分别在Middlebury数据集上进行匹配实验,结果表明,该算法平均性能提升了15.05%,明显优于传统自适应权重匹配算法。     

A flexible and configurable GPGPU stereo matching framework

This paper presents a practical framework that consists of various local to global stereo matching algorithms on a general purpose computing on graphics processing units platform. The flexible framework provides users with a selection of individual sub-algorithms in each step of stereo matching. The framework runs on either a central processing unit or graphic processing unit across three different platforms, including a smartphone, embedded board, and desktop. On the basis of the proposed framework, we investigate different combinations of stereo matching algorithms for specific use cases. Accordingly, we provide the framework’s quantitative speed and accuracy analysis evaluated on the widely used stereo dataset. In addition, this paper also addresses the parallelization strategy on an embedded graphics processing unit. The experimental results show that the proposed framework is capable of real-time and accurate depth estimation of stereo input in video graphics array resolution on an embedded graphics processing unit.

     

结合传播滤波的立体匹配算法研究

针对局部立体匹配算法在边缘处容易出现误匹配的问题,本文提出了一种结合权值传播进行代价聚合的局部立体匹配方法。首先采用基于颜色梯度的绝对差及Census方法构造了匹配代价函数;然后,引入传播滤波平滑匹配代价的同时保持视差空间图像边缘,与其他局部滤波器相比,该滤波器利用可传播的权值思想,不受传统局部算法窗口大小的影响;最后,通过左右一致性检查和无效视差值填充获得最终视差图。实验表明,该方法在Middlebury Stereo数据集上可获得精确结果,与Middlebury测试平台上的IGF、TSGO和Dog-Guided算法相比平均误差最低。     

Improving sub-pixel correspondence through upsampling

Many fundamental computer vision problems, including optical flow estimation and stereo matching, involve the key step of computing dense color matching among pixels. In this paper, we show that by merely upsampling, we can improve sub-pixel correspondence estimation. In addition, we identify the regularization bias problem and explore its relationship to image resolution. We propose a general upsampling framework to compute sub-pixel color matching for different computer vision problems. Various experiments were performed on motion estimation and stereo matching data. We are able to reduce errors by up to 30%, which would otherwise be very difficult to achieve through other conventional optimization methods.     

改进的跨尺度引导滤波立体匹配算法

作为双目三维重建中的关键步骤,双目立体匹配算法完成了从平面视觉到立体视觉的转化.但如何平衡双目立体匹配算法的运行速度和精度仍然是一个棘手的问题.本文针对现有的局部立体匹配算法在弱纹理、深度不连续等特定区域匹配精度低的问题,并同时考虑到算法实时性,提出了一种改进的跨多尺度引导滤波的立体匹配算法.首先融合AD和Census变换两种代价计算方法,然后采用基于跨尺度的引导滤波进行代价聚合,在进行视差计算时通过制定一个判断准则判断图像中每一个像素点的最小聚合代价对应的视差值是否可靠,当判断对应的视差值不可靠时,对像素点构建基于梯度相似性的自适应窗口,并基于自适应窗口修正该像素点对应的视差值.最后通过视差精化得到最终的视差图.在Middlebury测试平台上对标准立体图像对的实验结果表明,与传统基于引导滤波器的立体匹配算法相比具有更高的精度.     

RASCor: An associative hardware algorithm for real time stereo

Solving the stereo correspondence problem is an important step towards extraction of three dimensional structure of a scene from two or more images taken from distinct viewpoints. This paper presents a hardware solution to real time stereo matching problem using system of associative relations (SOAR) computational model. SOAR makes use of direction of derivatives for capturing pair-wise pixel interactions and uses pair-wise interactions to determine the underlying structure of associations. Pair-wise pixel interactions are defined over structures called Tokens. The hardware stereo matching solution developed in this study utilizes direction of derivatives to encode inter-pixel associations and implements a real time stereo matching solution. RASCor is realized on an FPGA platform and offers real time stereo matching using hardware parallelism and process pipelining. Real time stereo matching is achieved through hardware unit multiplicity and frequency scaling. The cost of hardware implementation is presented along with performance improvements from frequency scaling, unit multiplicity and search depth for stereo matching.     

Local algorithms on dense two-frame stereo matching

Stereo matching is a fundamental and crucial problem in computer vision. In the last decades, many researchers have been working on it and made great progress. Generally stereo algorithms can be classified into local methods and global methods. In this paper, the challenges of stereo matching are first introduced, and then we focus on local approaches which have simpler structures and higher efficiency than global ones. Local algorithms generally perform four steps: cost computation, cost aggregation, disparity computation and disparity refinement. Every step is deeply investigated, and most work focuses on cost aggregation. We studied most of the classical local methods and divide them into several classes. The classification well illustrates the development history of local stereo correspondence and shows the essence of local matching along with its important and difficult points. At the end we give the future development trend of local methods.     

立体视觉中的双目匹配方法研究

本文首先对已有的双目立体视觉方法进行分析和总结 ,并依所选取的匹配特征和匹配方法的不同而将其分为利用灰度图像区域间相似性、特征点相关、边界或二阶导数过零点、二值拉普拉斯图像匹配、校正透视形变、动态规划、利用区域分割的结果和立体视觉连续性原理的各种演绎等类 ;然后对用金字塔图匹配边界基元的双目立体视觉方法进行了重点探讨 ;本文还对利用基极线约束实现匹配进行详细分析与推导     

Stereo matching algorithm with guided filter and modified dynamic programming

Dense stereo correspondence is a challenging research problem in computer vision field. To address the poor accuracy behavior of stereo matching, we propose a novel stereo matching algorithm based on guided image filter and modified dynamic programming. Firstly, we suggest a combined matching cost by incorporating the absolute difference and improved color census transform (ICCT). Secondly, we use the guided image filter to filter the cost volume, which can aggregate the costs fast and efficiently. Then, in the disparity computing step, we design a modified dynamic programming algorithm, which can weaken the scanning line effect. At last, final disparity maps are gained after post-processing. The experimental results are evaluated on Middlebury Stereo Datasets, showing that our approach can achieve good results both in low texture and depth discontinuity areas with an average error rate of 5.14 % and strong robustness.     

A Taxonomy and Evaluation of Dense Two-Frame Stereo Correspondence Algorithms

Stereo matching is one of the most active research areas in computer vision. While a large number of algorithms for stereo correspondence have been developed, relatively little work has been done on characterizing their performance. In this paper, we present a taxonomy of dense, two-frame stereo methods. Our taxonomy is designed to assess the different components and design decisions made in individual stereo algorithms. Using this taxonomy, we compare existing stereo methods and present experiments evaluating the performance of many different variants. In order to establish a common software platform and a collection of data sets for easy evaluation, we have designed a stand-alone, flexible C++ implementation that enables the evaluation of individual components and that can easily be extended to include new algorithms. We have also produced several new multi-frame stereo data sets with ground truth and are making both the code and data sets available on the Web. Finally, we include a comparative evaluation of a large set of today's best-performing stereo algorithms.     

立体视觉中的三目匹配方法研究

本文提出了利用金字塔图传播匹配的三目基元立体视觉方法．该方法中匹配起始三 元组的选择建立在透视投影的可靠基础之上,而进行传播匹配可以加速计算,并避免了可能 的误匹配．传播匹配利用立体图像组在二值边界图像金字塔图中的结构相似性,而整个匹配 结果的考察与认定是通过传播率的比较实现的．该方法能够有效地克服双目立体视觉方法的 某些本质局限性,确保匹配结果精确可靠．