Affine hull based target representation for visual tracking

Handling appearance variations is a challenging issue in visual tracking. Existing appearance models are usually built upon a linear combination of templates. With such kind of representation, accurate visual tracking is not desirable when heavy appearance variations are in presence. Under the framework of particle filtering, we propose a novel target representation for tracking. Namely, the target candidates are represented by affine combinations of a template set, which leads to better capability in describing unseen target appearances. Additionally, in order to adapt this representation to dynamic contexts across a video sequence, a novel template update scheme is presented. Different from conventional approaches, the scheme considers both the importance of one template to a target candidate in the current frame and the recentness of the template that is kept in the template set. Comprehensive experiments show that the proposed algorithm achieves superior performances in comparison with state-of-the-art works.     

A new MM algorithm for constrained estimation in the proportional hazards model

The constrained estimation in Cox’s model for the right-censored survival data is studied and the asymptotic properties of the constrained estimators are derived by using the Lagrangian method based on Karush–Kuhn–Tucker conditions. A novel minorization–maximization (MM) algorithm is developed for calculating the maximum likelihood estimates of the regression coefficients subject to box or linear inequality restrictions in the proportional hazards model. The first M-step of the proposed MM algorithm is to construct a surrogate function with a diagonal Hessian matrix, which can be reached by utilizing the convexity of the exponential function and the negative logarithm function. The second M-step is to maximize the surrogate function with a diagonal Hessian matrix subject to box constraints, which is equivalent to separately maximizing several one-dimensional concave functions with a lower bound and an upper bound constraint, resulting in an explicit solution via a median function. The ascent property of the proposed MM algorithm under constraints is theoretically justified. Standard error estimation is also presented via a non-parametric bootstrap approach. Simulation studies are performed to compare the estimations with and without constraints. Two real data sets are used to illustrate the proposed methods.     

Self‐similarity analysis of vehicle driver's electrodermal activity

This paper characterizes stress levels via a self‐similarity analysis of the electrodermal activity (EDA) collected in a real‐world driving context. To characterize the EDA richness over scales, the fractional Brownian motion (FBM) process and its corresponding exponent H, estimated via a wavelet‐based approach, are used. Specifically, an automatic scale range selection is proposed in order to detect the linearity in a log scale diagram. The procedure is applied to the EDA signals, from the open database drivedb, originally captured on the foot and the hand of the drivers during a real‐world driving experiment, designed to evoke different levels of arousal and stress. The estimated Hurst exponent H offers a distinction in stress levels when driving in highway versus city, with a reference to restful state of minimal stress level. Specifically, the estimated H values tend to decrease when the driving environmental complexity increases. In addition, the estimated H values on the foot EDA signals allow a better characterization of the driving task than that of hand EDA. The self‐similarity analysis was applied to various physiological signals in literature but not to the EDA so far, a signal which was found to correlate most with human affect. The proposed analysis could be useful in real‐time monitoring of stress levels in urban driving spaces, among other applications.     

基于双向位图的CSR大规模图存储优化

大数据时代,Graph500是评测超级计算机处理数据密集型应用能力的重要工具,E级验证系统的图遍历处理能力主要受限于内存空间和访存带宽,尤其是内存空间利用率直接决定了图的测试规模和测试性能.针对天河E级验证系统小内存特征,提出了基于双向位图的大规模图数据压缩存储方法(bidirectional-bitmap based CSR,Bi-CSR),Bi-CSR在CSR矩阵压缩的基础上引入行方向位图和列方向位图协同完成稀疏矩阵压缩存储,行方向位图主要负责行方向位图的压缩存储与索引,列方向位图除了进一步压缩图存储空间,还负责为顶点遍历向量并行优化提供加速空间.Bi-CSR大幅度减少了稀疏矩阵存储空间.面向天河E级验证系统,当图输入规模为237时,Graph500的图存储空间节约效率接近70%,全系统稳定测试性能为2.131E+12TEPS,性能最大加速比超过100倍.     

基于混合远景和近景区域暗通道算法的图像去雾研究

为了提高图像去雾效果,提出混合远景和近景区域暗通道算法。首先建立远景和近景区域暗通道,为近景区域最小值通道和远景区域最小值通道之和,不同区域设置不同的调节系数;接着估计大气光值,通过混合暗通道远景区域的前50个像素点颜色值的平均值获得,增设校正系数;然后导向滤波,采用自适应滑动窗口优化透射率;最后给出了算法流程。实验仿真显示所提算法去雾图像视觉效果较好,定量分析指标对比度、可见边对比度值较优。     

一种灰度化混合法在集装箱箱号识别中的运用

研究一种基于机器视觉的集装箱箱号识别方法。对于集装箱彩色图像预处理过程中的灰度化方法，传统的灰度化算法不能有效弥补图像中污损或其他信息缺失的部分，因此，本文提出使用主成分分析法（PCA）结合贝叶斯阈值估计灰度变化率的混合法对图像的灰度化进行优化，可以在判断图像中某一点灰度值与周围相邻像素点的灰度值的变化率后，弥补缺失信息，有效确定边缘特征，从而使后续的字符识别准确率大大提高。最后使用该算法模型设计实现一套用于港口集装箱的智能检测系统。经过Matlab实验验证，在对50幅港口集装箱箱号图像的识别中，通过使用本文提出的混合灰度化方法，与普通的均值法和加权平均法的灰度化方法相比，准确率更高，其中单一字符准确率可达96%，箱号准确率可达92%。     

基于颜色四通道及空间金字塔的鱼类图像检索

随着计算机视觉技术在海洋水产领域中的应用不断加深，鱼类图像检索在渔业资源调查、鱼类行为学分析等方面发挥了巨大的作用。通过研究发现，鱼类图像的背景信息会对鱼类图像检索造成极大干扰，而且鱼类图像中颜色、纹理、形状等特征由于空间位置信息的缺乏而使检索的准确率不高。为解决以上问题，提出了一种新的基于颜色四通道及空间金字塔的鱼类图像检索算法。首先，提取视觉显著性图将鱼类图像的前景和背景分开，从而减少图像背景对检索的干扰；其次，为了使图像特征包含一定的空间位置信息，利用空间金字塔的理论对图像进行分割，在此基础上，将图像转为HSVG四通道图并提取SURF特征；；最后，得到检索结果。为验证所提算法的有效性，在QUT_fish_data数据集和DLOU_fish_data数据集上对算法的查全率、查准率与经典的HSVG算法和显著性分块算法进行对比：在两个数据集上查准率分别比传统的HSVG算法最多分别提高12%和5%，查全率最多分别提高7%和22%；比传统的显著性分块算法查准率最多分别提高15%和5%，查全率最多分别提高36%和22%；从而证明所提算法是有效的，能有效提升鱼类图像的检索效果。     

基于限定PSO初始化空间的随机最大似然算法

针对随机最大似然算法（SML）在波达方位（DOA）估计中由于多维非线性优化导致计算复杂度大的问题，提出一种限定粒子群（PSO）算法搜索空间的SML算法。该算法克服了一个缺陷，即在采用ESPRIT算法限定PSO初始化空间时，在阵列结构是非均匀线性阵列而且信号是相干信号时ESPRIT算法不能直接处理信号，且需要采用一组预处理技术，这增加了算法计算的复杂度。提出的算法的关键之处在于采用假设技术确定初始化点来代替ESPRIT算法的解，结合克拉美罗界（CRB）确定PSO算法的初始化解空间。这一方法不必再采用预处理技术，且利用限定PSO初始化空间的算法大大降低了SML算法的计算复杂度。实验结果表明，提出的算法为相干情况和非相干情况都提供了相当好的初始值。最后，将该算法与许多现有算法进行比较，验证提出算法的有效性和准确性。