基于Hu矩和纹理特征结合的人体异常行为识别

为了提高人体异常行为识别的准确率,采用了一种将多特征结合的异常行为识别算法,主要包括对步行、快跑、慢跑、拳击、双手挥舞、鼓掌六种异常行为进行识别。首先从视频流中提取出人体轮廓,然后从所得的轮廓中提取Hu矩特征与纹理特征。最后通过模板匹配的方法,采用马氏距离度量所需识别的当前行为特征向量与标准模板行为的特征向量之间的相似性,并通过设置相应的阈值判定该行为所属类别。实验证明,该方法比提取单一特征的方法识别率高,且具有一定的实用价值。     

Action detection by double hierarchical multi-structure space–time statistical matching model

Aimed at the complex information in videos and low detection efficiency, an actions detection model based on neighboring Gaussian structure and 3D LARK features is put forward. We exploit a double hierarchical multi-structure space–time statistical matching model (DMSM) in temporal action localization. First, a neighboring Gaussian structure is presented to describe the multi-scale structural relationship. Then, a space–time statistical matching method is proposed to achieve two similarity matrices on both large and small scales, which combines double hierarchical structural constraints in model by both the neighboring Gaussian structure and the 3D LARK local structure. Finally, the double hierarchical similarity is fused and analyzed to detect actions. Besides, the multi-scale composite template extends the model application into multi-view. Experimental results of DMSM on the complex visual tracker benchmark data sets and THUMOS 2014 data sets show the promising performance. Compared with other state-of-the-art algorithm, DMSM achieves superior performances.     

Moving object detection based on improved VIBE and graph cut optimization

Extracting foreground moving objects from video sequences is an important task and also a hot topic in computer vision and image processing. Segmentation results can be used in many object-based video applications such as object-based video coding, content-based video retrieval, intelligent video surveillance and video-based human–computer interaction. In this paper, we present a novel moving object detection method based on improved VIBE and graph cut method from monocular video sequences. Firstly, perform moving object detection for the current frame based on improved VIBE method to extract the background and foreground information; then obtain the clusters of foreground and background respectively using mean shift clustering on the background and foreground information; Third, initialize the S/T Network with corresponding image pixels as nodes (except S/T node); calculate the data and smoothness term of graph; finally, use max flow/minimum cut to segmentation S/T network to extract the motion objects. Experimental results on indoor and outdoor videos demonstrate the efficiency of our proposed method.     

结合SIFT算法的视频场景突变检测

视频场景变化检测对于视频的标注以及语义检索具有非常重要的作用。本文提出了一种结合SIFT(Scale Invariant Feature Transformation)特征点提取的场景变化检测算法。首先利用SIFT算法分别提取出视频前后帧的特征点并分别统计其数量,然后对视频前后帧进行图像匹配,统计匹配上的特征点数量,最后将该帧的匹配特征点数量与该帧前一帧的特征点数量做比值,从而通过该比值判断场景变化情况。实验结果表明,视频场景突变检测率平均可以达到95.79%。本算法可以在视频帧进行图像匹配的过程中对场景的变化情况进行判断,因此该算法不仅应用范围较广,还可以保证场景变化检测的精度,仿真结果证明了算法的有效性。     

基于特征匹配的电子稳像优化技术

针对特征匹配的电子稳像算法存在特征点提取耗时多、准确度不高以及特征点误匹配等问题,从特征点提取及特征点匹配两个方面采取了优化措施,采用基于快速分割特征点提取算法更快地提取特征点,并把特征点定位准确度提高到亚像素级;引入随机采样一致性算法消除误匹配的特征点对,减少场景中局部运动的影响,提高了全局运动参量的估计准确性.实验结果表明,经优化技术处理后,视频稳像准确度高、速度快,能够在存在平移、缩放、旋转及局部有相对运动等情况时保持良好的视频稳像效果.     

基于特征匹配的电子稳像优化技术

针对特征匹配的电子稳像算法存在特征点提取耗时多、准确度不高以及特征点误匹配等问题,从特征点提取及特征点匹配两个方面采取了优化措施,采用基于快速分割特征点提取算法更快地提取特征点,并把特征点定位准确度提高到亚像素级;引入随机采样一致性算法消除误匹配的特征点对,减少场景中局部运动的影响,提高了全局运动参量的估计准确性.实验结果表明,经优化技术处理后,视频稳像准确度高、速度快,能够在存在平移、缩放、旋转及局部有相对运动等情况时保持良好的视频稳像效果.     

鲁棒的车载红外视频电子稳像算法

针对车辆行进中,红外热像仪拍摄的视频序列存在复杂的随机抖动,提出基于BRISK特征点匹配的运动估计算法,计算出高精度全局运动矢量,同时对于特征点匹配时出现误匹配及场景中存在前景运动物体的情况,采用模糊聚类法分离全局运动和局部运动,提高了算法的鲁棒性。提出了基于Kalman粒子滤波算法,有效实现了复杂扫描运动和随机抖动的分离,并利用双线性插值法进行图像补偿。采用快速图像拼接法进行未定义区域处理,实现了图像全景输出。还利用车载红外热像仪实际拍摄的红外视频进行了稳像实验。实验结果表明,视频序列获得了很好的稳像效果,能够满足实际应用要求。     

视频文字信息检查工具的设计与实现

针对某些特定视频中,画面文字信息经常包含较为敏感文字信息,导致信息泄露,设计实现了一种视频画面中的文字的检测识别系统,对视频画面中的文字标语、文字条幅,新闻画面中的文字导语等信息进行识别与比对。采用基于双阈值的视频镜头分割算法,根据颜色直方图信息提取关键帧,采用最大稳定极值区域算法提取图像中稳定区域,通过聚类和级联分类器实现文字区域提取,最后将文字区域分割后进行OCR识别,实验表明,设计的系统针对复杂背景中的文字能够达到较高的检测识别率。     

Point pattern matching based on kernel partial least squares

Point pattern matching is an essential step in many image processing applications.This letter investigates the spectral approaches of point pattern matching,and presents a spectral feature matching algorithm based on kernel partial least squares(KPLS).Given the feature points of two images,we define position similarity matrices for the reference and sensed images,and extract the pattern vectors from the matrices using KPLS,which indicate the geometric distribution and the inner relationships of the feature points.Feature points matching are done using the bipartite graph matching method.Experiments conducted on both synthetic and real-world data demonstrate the robustness and invariance of the algorithm.     

一种鲁棒的目标跟踪方法

针对传统特征光流场跟踪方法中由于误差积累和错误匹配而导致的特征点丢失问题,基于一种新的Harris-SIFT特征点表示方法,提出基于预测帧与关键帧的算法框架,实现了光流场运动估计与局部特征识别相结合的目标跟踪方法.预测帧利用塔式分解和递归算法计算特征点的光流场运动矢量,使用运动矢量直方图获取目标的运动矢量,并剔除误匹配点;当特征点数量小于5个时,关键帧使用Harris-SIFT特征点进行局部特征匹配,利用仿射模型对目标精确定位及姿态修正.实验结果表明,本方法对视频序列中的纹理特征目标跟踪的鲁棒性较好,在背景复杂、目标遮挡或暂时丢失情况下,仍可以继续完成目标的可靠跟踪.     

基于多模板子空间的支持向量机多类自然形状识别

自然物体的检测与识别是机器视觉以及模式识别的重要任务。由于自然物体形状的多样性与柔性以及视觉判别的复杂性,使基于计算机的自然形状物体的准确检测与识别变得比较困难。提出了基于多模板子空间的支持向量机(SVM)多类自然形状识别方法。利用广义Hough变换表示自然形状物体轮廓,针对每个类别通过训练得到多个匹配模板;检测时利用多模板最近邻相关匹配进行粗检测,使用支持向量机进行分类。在相关匹配限定的子空间内收集训练样本,有效地降低了训练样本数目。实验结果证明所提出的自然形状检测与识别方法是十分有效的,大大改进了经典检测算法的检测效果以及自动化程度。     

复Ginzburg-Landau方程时空混沌的网络同步与参量辨识

研究了参量未知的时空混沌系统构成复杂网络的同步与参量辨识问题. 设计的参量辨识律可以有效地辨识复杂网络中所有节点时空混沌系统中的未知参量. 基于稳定性定理, 通过构造适当的Lyapunov函数, 确定了网络完全同步的条件. 以参量未知的一维复Ginzburg-Landau方程作为网络节点为例, 通过仿真模拟检验了参量辨识律以及同步方法的有效性.     

Infrared small moving target detection method based on graph matching

We apply graph matching method to detect infrared small moving targets using image sequences. Candidates （interest points） detected in the first frame form one graph and the same candidates in the last frame form another one. The real moving targets are extracted by matching these two graphs. Experimental results demonstrate that the proposed method is robust and efficient to the translation and rotation of the background.     

多特征融合匹配的多目标跟踪

针对复杂背景的视频图像多目标跟踪遮挡问题,提出了一种多特征融合匹配的多目标跟踪方法。基于自适应高斯混合背景模型重构和更新背景策略,实现当前帧背景减除和多目标检测;采用目标的颜色特征、质心位置、运动速度等特征进行融合匹配跟踪;最后,通过区域辅助判定策略将场景下的目标状态分为目标进入场景、目标退出场景、目标暂消、目标重现、目标融合和分裂5种状态,用质心预测方法和遮挡因子辅助匹配来提高匹配正确率。仿真实验结果表明：采用该方法跟踪同一目标和不同目标的相似度平均值分别为0.949 71和0.505 73,优于单一颜色特征信息匹配;目标遮挡结束后重新匹配相似度为0.972 83,实现了复杂背景下具有表面相似性的多目标实时跟踪。     

Holovideo: Real-time 3D range video encoding and decoding on GPU

We present a 3D video-encoding technique called Holovideo that is capable of encoding high-resolution 3D videos into standard 2D videos, and then decoding the 2D videos back into 3D rapidly without significant loss of quality. Due to the nature of the algorithm, 2D video compression such as JPEG encoding with QuickTime Run Length Encoding (QTRLE) can be applied with little quality loss, resulting in an effective way to store 3D video at very small file sizes. We found that under a compression ratio of 134:1, Holovideo to OBJ file format, the 3D geometry quality drops at a negligible level. Several sets of 3D videos were captured using a structured light scanner, compressed using the Holovideo codec, and then uncompressed and displayed to demonstrate the effectiveness of the codec. With the use of OpenGL Shaders (GLSL), the 3D video codec can encode and decode in realtime. We demonstrated that for a video size of 512×512, the decoding speed is 28 frames per second (FPS) with a laptop computer using an embedded NVIDIA GeForce 9400 m graphics processing unit (GPU). Encoding can be done with this same setup at 18 FPS, making this technology suitable for applications such as interactive 3D video games and 3D video conferencing.     

Bottom-up design and assembly with superatomic building blocks

Constructing specific structures from the bottom up with artificial units is an important interdisciplinary topic involving physics, chemistry, materials, and so on. In this work, we theoretically demonstrated the feasibility of using superatoms as building blocks to assemble a complex at atomic-level precision. By using a series of actinide-based endohedral metallofullerene (EMF) superatoms that can form one, two, three and four chemical bonds, a planar complex with intra- and inter-molecular interactions was assembled on the Au(111) surface. This complex is composed of two parts, containing ten and eight superatoms, respectively. The electronic structure analysis shows that the electron density inside each part is connected and the closed-shell electronic arrangement system is designed. There is also an obvious van der Waals boundary by physical adsorption between the two parts, and a stable complex is formed. Since this complex is realized by the first-principles calculations of quantum mechanics, our results help not only achieve atomic-level precision construction with artificial superatomic units but also maintain atomic-level functional properties.     

Dynamic photoelastic analysis by optical heterodyne polarimetry

This paper describes the mapping of the spatiotemporal principal stress distribution evolved with time in an epoxy photoelastic sample. In the optical heterodyne polarimeter exploited, the signal beam of light transmitted by the sample under continuously loaded condition is photomixed with the local oscillator beam of light made up of orthogonal linearly polarized two-frequency components. Every pixel of a MOS video camera used generates a beat photocurrent that possesses the two orthogonal field components of the elliptically polarized signal beam. The spatiotemporal principal stress distributions can be uniquely determined simultaneously and independently from these two orthogonal field components, and are successfully mapped in a time-sequential form. The spatial and temporal resolutions in the maps are 0.18 mm and 2.9 ms, respectively.     

Hand vein recognition based on the connection lines of reference point and feature point

According to the essential characters of the image topology, a new hand vein recognition algorithm based on the connection lines of reference point and feature points is proposed. In this method, the intersection points and the endpoints of the vein image are used as feature points. After the intersection points and the endpoints selected as feature points, the reference point for image matching are extracted from these points. The relative distances between the reference point and the feature points and the angles between the adjacent connections of the reference point and feature points are calculated and used as recognition features. Finally these two features are combined for hand vein recognition. This method can effectively overcome the influence on the recognition results caused by image translation and rotation. Experimental results show that the proposed algorithm is able to achieve hand vein recognition reliably and quickly.     

Chimera states in networks of logistic maps with hierarchical connectivities

Chimera states are complex spatiotemporal patterns consisting of coexisting domains of coherence and incoherence. We study networks of nonlocally coupled logistic maps and analyze systematically how the dilution of the network links influences the appearance of chimera patterns. The network connectivities are constructed using an iterative Cantor algorithm to generate fractal (hierarchical) connectivities. Increasing the hierarchical level of iteration, we compare the resulting spatiotemporal patterns. We demonstrate that a high clustering coefficient and symmetry of the base pattern promotes chimera states, and asymmetric connectivities result in complex nested chimera patterns.