一种抗遮挡的多运动目标跟踪改进算法

由于传统空间欧氏距离最短法难以解决遮挡问题,提出一种固定场景下抗遮挡的对多个运动目标进行实时检测和跟踪的算法.在分析传统帧差算法的优缺点的基础上对其进行了改进,引入空间滤波和区域填充.介绍了传统空间欧氏距离最短法,分析了它的缺点.用带状态参量的空间欧氏距离最短法对每个视频运动目标质心进行关联,监测每个视频运动对象的运动状态、运动轨迹.通过实验证明,该方法在改进传统欧式距离最短法的基础上,能实时有效得跟踪运动目标.     

Displacement Following of Hidden Objects in a Video Sequence

In a video sequence, computing the motion of an object requires the continuity of the apparent velocity field. This property does not hold when the object is hidden by an occlusion during its motion. The minimization of an energy functional leads to a simple algorithm which allows the recovery of the most likely trajectory of the occluded object from optical flow data at the border of the occlusion. Optical flow used for developing our method is an improvement on any variational technique of computing it. This improvement is based on a multichannel segmentation.     

Regulation and Release of Vasoactive Endoglin by Brain Endothelium in Response to Hypoxia/Reoxygenation in Stroke

In large vessel occlusion stroke, recanalization to restore cerebral perfusion is essential but not necessarily sufficient for a favorable outcome. Paradoxically, in some patients, reperfusion carries the risk of increased tissue damage and cerebral hemorrhage. Experimental and clinical data suggest that endothelial cells, representing the interface for detrimental platelet and leukocyte responses, likely play a crucial role in the phenomenon referred to as ischemia/reperfusion (I/R)-injury, but the mechanisms are unknown. We aimed to determine the role of endoglin in cerebral I/R-injury; endoglin is a membrane-bound protein abundantly expressed by endothelial cells that has previously been shown to be involved in the maintenance of vascular homeostasis. We investigated the expression of membranous endoglin (using Western blotting and RT-PCR) and the generation of soluble endoglin (using an enzyme-linked immunosorbent assay of cell culture supernatants) after hypoxia and subsequent reoxygenation in human non-immortalized brain endothelial cells. To validate these in vitro data, we additionally examined endoglin expression in an intraluminal monofilament model of permanent and transient middle cerebral artery occlusion in mice. Subsequently, the effects of recombinant human soluble endoglin were assessed by label-free impedance-based measurement of endothelial monolayer integrity (using the xCELLigence DP system) and immunocytochemistry. Endoglin expression is highly inducible by hypoxia in human brain endothelial monolayers in vitro, and subsequent reoxygenation induced its shedding. These findings were corroborated in mice during MCAO; an upregulation of endoglin was displayed in the infarcted hemispheres under occlusion, whereas endoglin expression was significantly diminished after transient MCAO, which is indicative of shedding. Of note is the finding that soluble endoglin induced an inflammatory phenotype in endothelial monolayers. The treatment of HBMEC with endoglin resulted in a decrease in transendothelial resistance and the downregulation of VE-cadherin. Our data establish a novel mechanism in which hypoxia triggers the initial endothelial upregulation of endoglin and subsequent reoxygenation triggers its release as a vasoactive mediator that, when rinsed into adjacent vascular beds after recanalization, can contribute to cerebral reperfusion injury.     

基于shader实现的光线透射效果

真实的模拟光线的透射效果是实时计算机图形学中一个非常令人关注的内容.对于这种自然现象的模拟所要达到的两个目标是真实性和实时性.为了达到这一要求,介绍了一种能够实时并且真实的模拟这种自然现象的算法,在Microsoft Visual Studio开发环境下,利用DirectX9.0软件包进行3D开发,并结合shader技术编程实现,运行的结果表明此算法是可行的和高效的.     

多窗口目标跟踪算法

目标在复杂环境下运动,特征会受到光照、摄像机角度等因素影响,并且目标可能会被自身其他部分或者场景物体遮挡,为此,提出一种多窗口概念的视频目标跟踪算法,基本思想就是将目标用多个不同的窗口表示,每一个窗口对应一个跟踪器.这些窗口之间可以存在重叠,也可以包含被跟踪目标上下文信息,如在人脸跟踪中,有些跟踪窗口可以包含颈部信息、上衣颜色信息.然后利用这些窗口之间的位置关系以及在后继跟踪过程中的相似度来估计目标的真实位置.通过实验可以看出,提出的多窗口均值移动跟踪算法可以提高系统在遮挡、光照变化等复杂条件下的跟踪性能.     

增强现实中基于自然特征的实时跟踪方法

在增强现实系统的复杂场景中,对目标的实时跟踪受到场景中诸多因素的制约,导致实时跟踪方法效率低且不准确,为此提出一种基于自然特征的实时跟踪方法。设计了一种螺旋分割模型,对捕获的图像进行螺旋分割,利用SURF算法在分割子块中提取特征点,并进行匹配。在对目标进行跟踪定位时,利用前一帧来预测当前帧目标出现的位置,以减少SURF算法的扫描区域,加速系统运算效率。实验中分别对场景光线强弱、视点和仿射变化以及目标被部分遮挡等不同情况进行测试,该方法均表现出较高的跟踪效率。     

系统振荡又发生短路故障时距离保护的动作行为分析

对几种常见的距离保护振荡闭锁原理进行简要分析,指出各种原理的优缺点,并提出今后努力的方向。     

多尺度相似性迭代查找的可靠双目视差估计

目的双目视差估计可以实现稠密的深度估计,因而具有重要研究价值。而视差估计和光流估计两个任务之间具有相似性,在两种任务之间可以互相借鉴并启迪新算法。受光流估计高效算法RAFT(recurrent all-pairs field transforms)的启发,本文提出采用单、双边多尺度相似性迭代查找的方法实现高精度的双目视差估计。针对方法在不同区域估计精度和置信度不一致的问题,提出了左右图像视差估计一致性检测提取可靠估计区域的方法。方法采用金字塔池化模块、跳层连接和残差结构的特征网络提取具有强表征能力的表示向量,采用向量内积表示像素间的相似性,通过平均池化得到多尺度的相似量,第0次迭代集成初始视差量,根据初始视差单方向向左查找多尺度的相似性得到的大视野相似量和上下文3种信息,而其他次迭代集成更新的视差估计量,根据估计视差双向查找多尺度的相似性得到的大视野相似量和上下文3种信息,集成信息通过第0次更新的卷积循环神经网络和其他次更新共享的卷积循环神经网络迭代输出视差的更新量,多次迭代得到最终的视差估计值。之后,通过对输入左、右图像反序和左右翻转估计右图视差,对比左、右图匹配点视差差值的绝对值和...     

基于统计分类的多行人跟踪算法

为了使视频监控系统具备多行人自动跟踪能力,提出了一种基于统计分类的多行人跟踪算法。首先采用了背景差和Canny算子准确提取出运动区域,然后利用特征匹配实现帧间目标的对应,最后提出一种利用人体表面颜色空间分布信息的统计分类方法,完成合并或遮挡时多个行人的分离和单独跟踪。实验表明,在相互遮挡时,该算法也能有效地分割和跟踪多个行人,适于智能视觉监控应用。     

部分遮挡目标的稳健局部特征点提取方法

部分遮挡目标的特征提取是目标检测中的难点问题。针对这一问题,提出一种基于稳健性度量统计分析的部分遮挡目标稳健局部特征点提取方法,该方法建立在目标图像训练集的局部特征点逐步提纯基础上。它首先对目标在不同条件下获取的图像集分别应用SIFT提取相应局部特征点;然后基于统计分析对局部特征点进行粗提取,再通过计算置信度进一步筛选出较稳健的局部特征点;最后分析局部特征点的空间分布等因素提取出最稳健的局部特征点集。实验结果表明,该方法耗时仅为SIFT检测算法的30%,并且保证了检测精度。