Shape Context算法的CUDA并行化实现

研究了Shape Context形状识别串行算法,在此基础上针对CUDA平台多处理器流水线特性,将Shape Context算法进行适应CUDA特性的并行化改造,使得算法并行化后具有更高的执行精度,并达到了130倍以上的加速比,较大程度满足人脸识别海量数据的实时处理需求.     

基于多帧杆号字符识别的接触网检测系统

针对传统电气化铁路接触网检测在稳定性和准确性方面的不足,提出了以自动 识别的杆号作为接触网杆定位和图像索引检测依据的接触网自动检测系统。该系统介绍了多帧图像杆号识别过程,比较了形状上下文（Shape context,SC）算法、角点典型形状上下文特征（Corner representative shape context,CRSC ）算法和重心形状上下文（Center shape context,CSC）算法,确定了将 CSC算法作为杆号识别的算法。实验结果表明,该算法具有实时性好、可靠性高等优点,能够满足时速300 km左 右的接触网实时检测要求,为电气化铁路定位检测提供了一种稳定性好、检测速度快的方法。     

基于Context模型的小波变换阈值自适应图像去噪

根据噪声和信号的小波系数在不同分解尺度、不同方向上高频系数的分布不同,结合Context模型,提出基于Context模型的小波变换阈值自适应图像去噪算法。该算法通过对不同尺度和方向的小波分解系数应用不同的阈值方法进行去噪。实验表明,方法能较好地去除图像噪声和保留图像边缘细节信息,在提高去噪图像信噪比值和改善视觉效果方面都表现出了良好的性能。     

一种改进的类人足球机器人彩色目标识别算法

针对类人足球机器人视觉需求,提出一种结合区域生长和基于形状判别的阈值自适应更新的彩色目标识别算法。该算法在HSI空间基于S分量把图像分为高饱和区域和低饱和区域,在高饱和区域基于H分量采用区域生长算法识别目标;通过目标形状判别自适应更新阈值,并用新阈值更新区域生长中原来的阈值,以稳定准确地识别彩色目标。在类人足球机器人系统中的成功应用表明,该算法能在不同光照条件下稳定地识别出彩色目标,对光照环境有良好的适应性和鲁棒性,具有良好的识别效果。     

基于激光传感器的稻种轮廓形状测量方法研究

稻种轮廓形状测量是稻种基于形状特征识别的前提条件.以激光传感器采集到的稻种表面点云为基础,提出了一种基于主平面剖视图投影的稻种轮廓形状测量方法.通过主成份分析法校正点云模型的坐标系,利用主平面投影法获取稻种的六视图投影模型,应用Alpha Shape算法提取投影点云模型的边缘轮廓,并计算稻种投影模型的周长、面积、长、宽、长宽比、最大半径、最小半径、半径比、圆形度等9个形状特征参数.选取大华香糯、豫粳6号、新稻10号3大类,每种各200粒稻种作为试验样品,采用径向基函数(RBF)神经网络模型对提取到的形状特征进行训练识别,识别率分别达到96%,94%,98%.结果表明:基于激光传感器的稻种轮廓形状测量方法能够较好地适用到稻种识别中.     

铁路接触网支柱的图像序列自适应识别方法

接触网支柱数字化管理是电气化铁路运维的关键环节,基于移动视频建立接触网支柱数字台账是高效、经济、便捷的技术手段。为实现对于移动视频图像序列中接触网支柱杆号的精准识别,提出了一种基于区域相关和改进SVTR网络的接触网支柱识别算法。针对视频图像中接触网支柱区域重叠、结构模式复杂的特点,采用了YOLO v4网络对单帧图像中支柱区域和号牌标识区域分别进行检测,并通过测算交叠区域来获得距观察点最近的杆位和对应的号牌区域。此外,针对接触网杆号牌尺度多样性和字符变长的问题,在杆号文字识别问题中采用了SVTR-tiny网络,并进一步引入迁移学习方法增强模型对于复杂杆号的识别精度和对于不同线路场景的泛化性能。通过在实际高铁线路采集的移动视频数据集上进行测试,结果表明算法在移动视频中视野最近杆位杆号区域的定位检出率可达98.01%,杆号文本的识别准确率达到96.13%,适用于我国高速铁路主要干线建设配套的接触网支柱结构。     

基于内距离形状上下文的跌倒检测方法

在全球社会老龄化的大背景下,老年人的身体健康状况和晚年生活质量需要更多的关注。跌倒在老年人群中发生率高并且带来的后果比较严重。文中提出一种应用于家庭场景的基于Inner-Distance形状上下文（ Inner-Distance Shape Context,IDSC）的跌倒检测方法。该方法通过Inner-Distance形状上下文获得视频帧前景形状的描述信息,使用形状匹配方法对视频序列中人体形状变化进行量化。对形变量化信息使用动态时间规整（ Dynamic Time Warping,DTW）方法实现跌倒行为的判定。实验结果表明该方法可有效、快速地判断跌倒,提取结果具有较好的查准率和查全率。     

基于多尺度轮廓段的形状特征提取与识别

形状识别是计算机视觉与模式识别领域的重要研究内容。形状的特征选取与描述是形状识别的研究热点。针对现有识别方法的不足，提出一种通过对不同长度轮廓段进行描述，进行特征提取的方法。对每个形状均在6种尺度下进行特征提取，每种尺度选取5种轮廓段特征参数，实现了对形状的特征描述。在形状识别阶段，使用动态时间规整（DTW）算法度量形状描述子之间的匹配距离，实现形状识别。分别在Kimia99、Kimia216和MPEG-7数据库中进行算法验证，结果表明基于多尺度轮廓段的形状特征描述子具有旋转、缩放、平移和局部遮挡不变性，识别率优于现有算法。     

自适应稀疏编码融合的非刚性三维模型分类算法

针对传统的三维模型单一特征识别精度低问题,提出一种自适应稀疏编码融合的非刚性三维模型分类算法.首先提取模型的平均测地线特征、热核特征、形状直径函数特征,以构造互补的多特征形状描述;其次利用特征袋模型构造AGD-BoF, HKS-BoF, SDF-BoF特征向量,通过随机样本建立特征权值矩阵;最后利用特征权值矩阵与稀疏优化编码自适应融合,并采用Softmax分类算法实现非刚性模型的有效分类.通过在非刚性数据集SHREC10与SHREC11上的综合实验表明,自适应稀疏编码融合的分类算法具有更高的识别准确率及较强的鲁棒性.     

基于SURF和形状上下文的人脸匹配算法

针对人脸匹配在光照、姿态、表情等背景因素影响下匹配正确率低的问题,本文提出一种基于SURF(Speeded up Robust Features)和形状上下文(Shape Context, SC)的人脸图像匹配算法。在对图像进行人脸区域检测和重构积分图预处理的基础上,本文利用两次SURF算法提取人脸特征点并匹配,第一次用SURF进行粗匹配得到初始匹配集,并据此计算待匹配人脸图像间的尺度差和角度差,以此作为约束条件第二次进行SURF精匹配,以获得更多的匹配点对,最后采用不依赖位置信息的形状上下文算法剔除误匹配点对。在IMM和Georgia人脸数据库上与目前流行的人脸匹配方法进行实验对比,实验结果显示本文算法有效增加了匹配点对数目,并提高了人脸图像匹配正确率,具有更好的稳定性和鲁棒性。     

具有减少辨识参数的随机多变量极点配置自适应控制器

本文提出了一种新的限制输出个数减少随机多变量自适应控制中辨识参数的方法，并给出了减少辨识参数的极点配置自适应算法。虽然采用ｎ个输入１个输出的减少辨识参数的模型来设计控制器，但所提出的控制器能够保证被控系统的几个输出跟踪参考输入信号，仿真结果表明，所提出的方法是成功的。     

Robust adaptive pole placement for linear time-varying systems

This paper studies the robustness properties of a basic adaptive control algorithm with respect to plant parameter variation as well as modeling errors and bounded disturbances. The algorithm consists of a projected gradient estimator and a pole assignment controller. A unified method of proof is presented for robust stability of both discrete and continuous-time adaptively controlled time-varying systems. The robust performance of an adaptive pole assignment controller is also discussed     

间接自适应极点配置控制算法的跟踪能力研究

为了研究现有间接自适应极点配置控制算法的跟踪能力,提出了一般形式的间接自适应 极点配置控制算法.分析了这种算法用于已知和未知系统时的跟踪能力,并分别给出了渐近 跟踪参考输出的充要条件.利用上述结果,可以对现有间接算法的跟踪能力进行逐一研究.研 究表明,现有间接自适应极点配置控制算法的跟踪能力是有限的,它们至多能实现对一类参考 输出的渐近跟踪.     

Adaptive simplification of solution for support vector machine

SVM has been receiving increasing interest in areas ranging from its original application in pattern recognition to other applications such as regression estimation due to its remarkable generalization performance. Unfortunately, SVM is currently considerably slower in test phase caused by number of the support vectors, which has been a serious limitation for some applications. To overcome this problem, we proposed an adaptive algorithm named feature vectors selection (FVS) to select the feature vectors from the support vector solutions, which is based on the vector correlation principle and greedy algorithm. Through the adaptive algorithm, the sparsity of solution is improved and the time cost in testing is reduced. To select the number of the feature vectors adaptively by the requirements, the generalization and complexity trade-off can be directly controlled. The computer simulations on regression estimation and pattern recognition show that FVS is a promising algorithm to simplify the solution for support vector machine.     

On adaptive minimum variance regulation for non-minimum phase plants

Adaptive minimum variance control is applied to nonminimum phase plants augmented with adaptive compensators. The objective of the compensators is to achieve, asymptotically, a minimum phase property for the augmented plant. With this property, the minimum variance controller gives a stabilizing control signal. The scheme proposed has interpretations as an adaptive quadratic index minimizing procedure and also as an adaptive pole assignment algorithm. The advantage over other adaptive quadratic index minimizing procedures is simply that the weightings are made adaptive so as to guarantee convergence. One feature not shared by other adaptive pole assignment schemes is that they conveniently specialize to simple minimum variance schemes when, for these, control energy constraints are not violated. This paper presents only the discrete-time adaptive regulator version. Companion papers give a full global convergence theory and pre-processing augmentations for the stochastic tracking scheme versions.     

Radial basis function neural network-based face recognition using firefly algorithm

This paper presents an adaptive technique for obtaining centers of the hidden layer neurons of radial basis function neural network (RBFNN) for face recognition. The proposed technique uses firefly algorithm to obtain natural sub-clusters of training face images formed due to variations in pose, illumination, expression and occlusion, etc. Movement of fireflies in a hyper-dimensional input space is controlled by tuning the parameter gamma (γ) of firefly algorithm which plays an important role in maintaining the trade-off between effective search space exploration, firefly convergence, overall computational time and the recognition accuracy. The proposed technique is novel as it combines the advantages of evolutionary firefly algorithm and RBFNN in adaptive evolution of number and centers of hidden neurons. The strength of the proposed technique lies in its fast convergence, improved face recognition performance, reduced feature selection overhead and algorithm stability. The proposed technique is validated using benchmark face databases, namely ORL, Yale, AR and LFW. The average face recognition accuracies achieved using proposed algorithm for the above face databases outperform some of the existing techniques in face recognition.

     

Continuous-time weighted least-squares algorithm and its application to adaptive feedback control

A weighted least-squares algorithm is examined for continuous-time linear deterministic systems. The convergence of the algorithm is discussed from several viewpoints. The necessity of the persistently exciting condition is stressed for the exponential convergence of the algorithm. The need and the effect of introducing the normalizing factor into the algorithm are also considered in relation to the global stability of the adaptive feedback control. Adaptive pole placement control by means of an adaptive observer is considered as a typical example of the application of the algorithm.     

A flaw in the reduced-order behavior of a direct adaptive pole placer

This paper discusses and simulates an apparently fatal flaw in the behavior of the reduced-order application of a recently proposed direct adaptive pole placer. The algorithm can exhibit instability, in the output and input of the controlled plant despite the stability, of the prediction error driving the parameter adaption equations. This occurs when the order of each of the directly adjusted compensator elements is less than that of the plant and, correspondingly, the number of prespecified desired (nonzero) pole locations is less than twice the order of the plant, which arises from underestimating the plant order.     

车载命令词识别系统背景音乐的自适应抵消

在车载命令词识别系统中,背景音乐的播放降低了命令词识别率。而音乐信号因自相关矩阵特征值扩散度较大和谱平坦度较小在算法收敛速度上比语音信号慢,以至于传统的自适应抵消算法很难将音乐干扰消除干净,保证不了命令词识别率。为了解决这一问题,文中引入了预白化自适应滤波器来减小其自相关矩阵特征值扩散度和增大谱平坦度,并将此方法结合双滤波自适应算法,用来消除车内的背景音乐,以提高车载命令词识别系统的识别率。实验结果表明,经过背景音乐消除,命令词识别率有明显的提高,并且预白化也提高了识别率。     

多变量自适应极点配置的递推算法

本文提出一个递推计算控制器参的方法，得到具有两层迭代过程的多变量自适应极点配置算法，大幅度减少了在线计算量，同时在辩识算法中利用投影方法，证明了该算法的稳定性和收敛性。