基于鲁棒M-估计器的全局运动估计方法

提出一种改进的基于鲁棒M-估计器的全局运动估计方法,在图像像素残差上采用在统计上鲁棒的M-估计目标函数,引入简化的像素残差权值函数作为二值函数,改进了全局运动估计中去除噪声点的方法。实验结果表明,改进算法的运动精度高,对大多数的图像序列都有较好鲁棒性且计算量小。     

判别特征回归的鲁棒目标跟踪

目的 近年来,目标跟踪领域取得了很大进步,但是由于尺度变化,运动,形状畸变或者遮挡等造成的外观变化,仍然是目标跟踪中的一大挑战,因而有效的图像表达方法是提高目标跟踪鲁棒性的一个关键因素。方法 从中层视觉角度出发,首先对训练图像进行超像素分割,将得到特征向量集以及对应的置信值作为输入值,通过特征回归的方法建立目标跟踪中的判别外观模型,将跟踪图像的特征向量输入该模型,得到候选区域的置信值,从而高效地分离前景和背景,确定目标区域。结果 在公开数据集上进行跟踪实验。本文算法能较好地处理目标尺度变化、姿态变化、光照变化、形状畸变、遮挡等外观变化;和主流跟踪算法进行对比,本文算法在跟踪误差方面表现出色,在carScale、subway、tiger1视频中能取得最好结果,平均误差为12像素,3像素和21像素;和同类型的方法相比,本文算法在算法效率上表现出色,所有视频的跟踪效率均高于同类型算法,在carScale视频中的效率,是同类算法效率的32倍。结论 实验结果表明,本文目标跟踪算法具有高效性和鲁棒性,适用于目标发生外观变化时的目标跟踪问题。目前跟踪中只用了单一特征,未来考虑融合多特征来提升算法鲁棒性和准确度。     

具有不确定动态线性系统的鲁棒状态估计

本文研究了一类具有参数和噪声特性不确定线性系统的鲁棒状态估计问题。利用对策论思想,定义了能使不确定下最坏性能最好的极小极大鲁棒状态估计器,提出了一种简单的近似设计方法,即设计最坏对象的最优滤波器。给出了这种设计方法设计滤波器导致的性能误差边界,进一步指出当满足文中给出的鞍点条件时,最坏对象的最优滤波器就是极小极大鲁棒滤波器。     

基于多线索融合的鲁棒粒子滤波跟踪算法

为了提高目标特征的表达能力和跟踪的鲁棒性,提出基于多线索的目标跟踪算法。该算法分别从目标颜色特征和结构特征来考虑,在此基础上提出了融合公式,从而使目标在不同场景中都能自适应变化,以提高跟踪的精度和性能,最后,通过几组仿真实验对该算法进行了验证。实验结果表明,该算法对于部分遮挡等情况具有良好的鲁棒性和跟踪精度。     

一种快速鲁棒运动估计方法

经典的相位法易受高频噪声影响,且计算量大,难以满足实时性要求。提出了基于提升小波变换和离线预建立相关归一化系数与旋转角度对应关系数据库的快速鲁棒运动估计方法。试验结果表明,该方法在保持较高估计精度的情况下,具有较快的估计时间和较好的鲁棒性。     

基于快速鲁棒特征集合统计特征的图像分类方法

针对现有利用快速鲁棒特征(SURF)进行图像分类的方法中存在的效率低、正确率低的问题,提出一种利用图像SURF集合的统计特征进行图像分类的方法.该方法将SURF的各个维度及尺度信息视为各自独立的随机变量,并利用拉普拉斯响应区分不同数据.首先,获取图像的SURF向量集合;然后,分维度计算SURF向量集合的一阶中心绝对矩、带权一阶中心绝对矩等统计特征,并构建特征向量;最后,结合支持向量机(SVM)进行图像分类.在Corel 1K图像库上的实验结果表明,该方法查准率较SURF直方图方法和三通道Gabor纹理特征方法分别提高17.6%和5.4%.通过与HSV直方图特征进行高级特征融合,可获得良好的分类性能.与SURF直方图结合HSV直方图方法、三通道Gabor纹理特征结合HSV直方图方法、基于视觉词袋(BoVW)模型的多示例学习方法相比,查准率分别提高了5.2%,6.8%,3.2%.     

具有不确定噪声的连续广义线性系统的鲁棒状态估计

本文讨论了一类具有不确定噪声的连续广义线性系统的鲁棒状态估计问题，文章提供了一种比较实用的状态估计方法，针对噪声不确定性。文章采用对策论的基本原理，导出了 一种最小化不确定下最坏性能的极小极大鲁棒状态估计器。     

基于Huber M估计的鲁棒Cubature 卡尔曼滤波算法

Cubature卡尔曼滤波器(CKF)在非高斯噪声或统计特性未知时滤波精度将会下降甚至发散,为此提出了统计回归估计的鲁棒CKF算法.推导出线性化近似回归和直接非线性回归的鲁棒CKF算法,直接非线性回归克服了观测方程线性化近似带来的不足.具有混合高斯噪声的仿真实例比较了3种Cubature卡尔曼滤波器的滤波性能,结果表明这两种鲁棒CKF滤波精度及估计一致性明显优于CKF,直接非线性回归的CKF的鲁棒性更强,滤波性能更好.     

利用协变量调整控制混杂因子的鲁棒文本分类

针对目前很多文本分类方法很少控制混杂变量,且分类准确度对数据分布的鲁棒性较低的问题,提出一种基于协变量调整的文本分类方法.首先,假设文本分类中的混杂因子(变量)可在训练阶段观察到,但无法在测试阶段观察到;然后,以训练阶段的混杂因子为条件,在预测阶段计算出混杂因子的总和;最后,基于Pearl的协变量调整,通过控制混杂因子来观察文本特征和分类变量对分类器的精度影响.通过微博数据集和IMDB数据集验证所提方法的性能,实验结果表明,与其他方法相比,所提方法处理混杂关系时,可以得到更高的分类准确度,且对混杂变量具备鲁棒性.     

基于Huber的鲁棒高阶容积卡尔曼滤波算法

为提高随机变量非高斯分布时高阶容积卡尔曼滤波（High-degree Cubature Kalman Filter,HCKF）算法的鲁棒性,提出了一种基于Huber方法的鲁棒高阶容积卡尔曼滤波算法。从近似贝叶斯估计角度解释了Huber方法作用于卡尔曼滤波算法的本质是对新息进行截断平均,通过在现有滤波框架内利用Huber方法对观测量进行预处理,并将处理后的观测量进行标准的HCKF量测更新,实现了HCKF算法的鲁棒化。所提算法无需通过统计线性回归模型对系统的非线性量测模型进行近似,高阶容积变换的优势得到充分利用,从而在保持鲁棒性的前提下提高了算法的滤波精度。单变量非平稳增长模型和再入飞行器目标跟踪问题验证了该算法在鲁棒性和滤波精度方面的优势。     

Input estimation with multiple model for maneuvering target tracking

To increase the performance of maneuvering target tracking, an algorithm utilizing input estimation with multiple model based on two independent mode sets is suggested in this paper. The proposed algorithm consists of hypothesized multiple filters to estimate the unknown target acceleration and a test statistic developed from a modified version of the generalized likelihood ratio test to detect the maneuver onset time. An efficient algorithm for the target acceleration estimation is derived to reduce the computational burden of multiple model estimation. A numerical analysis is carried out to obtain the proper window length and the average delay of the algorithm. Performance of the proposed algorithm is evaluated by a series of simulation runs.     

Modified UBB-based input estimation method for nonlinear manoeuvring target tracking problem

The input detection and estimation methods in the manoeuvring target tracking (MTT) application need algorithms for manoeuvring detection and covariance resetting. This algorithm causes an improper delay in target states tracking. In this paper, for solving this problem, unknown but bounded approach for uncertainties modelling is used and a different state space model is developed. In this model, target acceleration is treated as an augmented state in the corresponding state equation. By using interval mathematics, the linearisation error is bounded by an ellipsoidal set and considered in the model development. In augmented state equations, the MTT problem converted to non-manoeuvring target tracking problem. Therefore, the set membership filter is rearranged and used for simultaneous target state and manoeuvre estimation. Furthermore, estimated convex set boundedness is analysed and an upper bound for the estimation error is calculated. The theoretical development of the proposed method is verified with numerical simulations, which contain examples of tracking various manoeuvring targets. The simulation result of the proposed method is compared with traditional input estimation methods. The comparison shows the acceptable performance of the proposed method in the simultaneous estimation of the target acceleration and state vector for the manoeuvring and non-manoeuvring scenarios.     

Joint target tracking and classification via RFS-based multiple model filtering

Firstly, a multiple model extension of the random finite set (RFS)-based single-target Bayesian filtering (STBF), referred as MM-STBF, is presented to accommodate the possible target maneuvering behavior in a straightforward manner. This paper is concerned with joint target tracking and classification (JTC) which are closely coupled. In particular, we take into account extraneous target-originated measurements which were not modeled in the existing JTC algorithms. Therefore, the main contribution is that the paper derives a new JTC algorithm based on the MM-STBF, i.e., MM-STBF–JTC. The MM-STBF–JTC is an optimal Bayesian solution, which can simultaneously accommodate unknown data association, miss-detection, clutter and several measurements originated from a target. The MM-STBF–JTC can reduce to a traditional JTC algorithm under some assumptions. The simulation results are provided to demonstrate the tracking and classification performance of the MM-STBF–JTC algorithm.     

对高速运动辐射源的实时高精度时差估计方法

在无源时差定位中,要求能实时获得目标到达测量站的TDOA参数,传统的互相关法需要大数据量、大运算量以及数据块处理,难以提取快速时变参数。提出了一种基于同伦延拓EKF的新方法,收敛速度快,估计精度达到CRLB下限,对每一个新的数据样点可直接更新TDOA参数,克服了一般EKF法中初始估计误差过大将导致滤波发散的问题。     

一种基于运动估计的红外目标跟踪方法

为解决红外目标跟踪中目标的交错、遮挡等问题,提出了一种新的基于运动估计的目标跟踪方法。建立目标的方向梯度-灰度直方图特征模型,该模型能较准确地刻画目标特征。使用最大后验概率指标在搜索区域进行目标匹配,该指标能很好地突出目标、抑制背景,并容易得到全局最优解。提出一种新的运动估计方法,即轨迹预测算法,对目标的运动进行较准确的估计。实验结果证明,该方法不仅计算复杂度低,而且能够较好地解决目标交错、遮挡等问题。     

基于分布估计算法的模糊分类方法

提出了基于分布估计算法的模糊分类建模方法,该方法基于Apriori原理生成初始模糊规则集,并且以匹茨堡型的二进制编码方式对模糊规则集编码,基于双变量相关的MIMIC (mutual information maximization for input clustering)分布估计算法从初始规则集中自动抽取模糊规则.通过在Iris,Pima,Wine这3个标准数据集的仿真实验表明,该方法比基于遗传算法的模糊分类器在准确率和解释性方面更有效.     

Phase detection based range estimation with a dual-band robust Chinese remainder theorem

Range estimation with multifrequency phases is a common practice in localization systems.The challenge of this method is the phase ambiguity.Some Chinese remainder theorem(CRT)based phase unwrapping algorithms have been proposed to solve the problem,where the wavelengths of the multifrequency signals need to be pair-wisely co-prime after they are divided by their greatest common divisor(gcd).This condition may limit the application in practice.In this paper,a novel way based on a dual-band robust CRT is presented to reconstruct the distance from dual-band wrapped phases,where the pair-wisely co-prime condition is not necessarily needed.As more wrapped phases are involved to reconstruct the distance,the method can significantly enlarge the reconstruction range compared to the single band solution.     

优化LeNet-5网络的多角度头部姿态估计方法

针对在受到部分遮挡或角度过大无法定位面部关键特征点的情况下,传统的头部姿态估计方法的准确率低或无法进行头部姿态估计的问题,提出了优化LeNet-5网络的多角度头部姿态估计方法.首先,通过对卷积神经网络(CNN)的深度、卷积核大小等进行优化来更好地捕捉图像的全局特征;然后,改进池化层,用卷积操作代替池化操作来增强网络的非...     

基于图像分类与多算法协作的目标跟踪算法

针对目标变化和背景环境的变化,提出了一种基于图像分类的多算法协作的目标跟踪算法,采用融入改进背景加权的尺度方向自适应均值漂移算法与快速压缩算法协作的方式。该算法根据图像变化原因不同将图像分为两类,图像全局变化和目标局部感兴趣区域的变化。对由光照、背景相似度和背景模糊引起的图像全局变化,采用快速压缩算法对目标进行跟踪;对由目标本身尺寸、旋转和遮挡引起的目标局部感兴趣区域变化,采用融入改进背景加权尺度方向自适应均值漂移算法对目标进行跟踪。该算法先对图像序列预处理分类,然后选择适合该对应图像变化特点的算法对目标进行跟踪。经实验验证,该算法较之其他流行目标跟踪算法具有更好的鲁棒性。     

Central-tendency estimation and nearest-estimate classification of event related potentials

Event related potentials (ERPs) are modeled as random vectors in order to determine multivariate central-tendency (C-T) estimates of ERPs such as the arithmetic mean, geometric mean, harmonic mean, median, tri-mean, trimmed-mean, and the Winsorized mean. Additionally, it is shown that the C-T estimates can be used to implement various forms of minimum-distance classifiers for individual channels and for single-channel heterogeneous, multi-channel homogeneous, and multi-channel heterogeneous-homogenous ERP classification through decision fusion. The study also focuses on answering the following related questions: (a) How do the C-T ERP estimates compare with each other? (b) How do the performances of nearest-estimate classifiers compare with each other? (c) For a given ERP channel, do the heterogeneous nearest-estimate classifiers offer complementary information for improving performance through decision fusion? (d) Do the homogeneous nearest-estimate classifiers of different channels offer complementary information for improving performance through decision fusion? (e) Can the performance be improved by fusing the decisions of all or a selected subset of the entire classifier ensemble? These questions are answered by designing estimation and classification experiments using real 6-channel ERPs. It is shown that although the operations to compute the vector C-T estimates can be quite different, the ERP estimates are similar with respect to their overall waveform shapes and peak latencies. Furthermore, the results of the classification experiments show that by fusing homogeneous nearest-estimate classifier decisions across multiple channels, the classification accuracy can be improved significantly when compared with the accuracies of individual channel classifiers.