一种新的大型四肢动物体姿态识别方法

文中以对鹿体姿态的识别为例，针对鹿体姿态的特点，提出了一种新的可应用于大型四肢动物体姿态识别的方法。首先采用数学形态学的方法对图像进行了处理，获得了鹿体图像南准确轮廓，然后分别用四个椭圆对所获得的图像中鹿的颈部、去除颈部的躯干以及四肢进行拟合．最后利用线性代数的知识分别求出这四个用以表征鹿体姿态信息的椭圆的长短轴长度和方向。     

基于三维人脸建模的多姿态人脸识别

待匹配人脸图像与原始图像存在姿态和光照的差异,是自动人脸识别的两个主要瓶颈问题.给出了采用三维人脸模型来解决人脸姿态的变化对人脸识别的影响问题.通过正侧面图像,利用B样条曲线与径向基函数相结合的方法进行三维人脸重建,生成三维人脸模型库.分别计算待匹配人脸图像的3个自由度,快速估计出人脸的姿态;结合待匹配人脸图像的姿态参数及三维人脸模型库,获得与待匹配图像相同姿态的三维人脸模型库中的二维人脸图像.最后完成了相同人脸姿态的二维人脸识别.实验结果证明,该方法无需复杂的设备、简单易行、识别时间短,是一种非常实用的解决人脸姿态问题的识别方法.     

基于姿态对齐的行人重识别方法

行人重识别是指根据输入的某个行人图片, 在视频监控网络中对该行人目标进行检索. 行人的姿态变化和监控场景的亮度变化是该任务的两个主要挑战. 针对行人的姿态变化问题, 本文首先对训练集中行人图片进行稠密 图像块采样获得图像块集合, 然后对每一个图像块提取其局部表观空间特征, 最后在此特征集上聚类得到通用的行人部件字典. 由于该部件字典编码了行人的部件信息, 因此通过该字典内的每一个码元可以建立两幅行人图像中特定图像块之间的对应关系. 将两幅行人图片的图像块集合分别向部件字典投影, 可以获得2幅行人图片姿态对齐后的图像块序列. 针对监控场景的亮度变化问题, 本文在姿态对齐后的图像块上分别提取4种颜色描述子, 并将不同颜色描述子下的图像块相似性进行分数级组合以获得更好的亮度不变性. 其中不同颜色描述子之间的组合系数通过结构化输出支持向量机学习得到. 在常用的视点不变行人重识别(viewpoint invariant pedestrian recognition,VIPeR)数据集上的实验结果表明, 该方法在存在行人姿态变化和场景亮度变化干扰时获得了较好的行人重识别效果.     

基于姿态对齐的行人重识别方法（英文）

行人重识别是指根据输入的某个行人图片,在视频监控网络中对该行人目标进行检索.行人的姿态变化和监控场景的亮度变化是该任务的两个主要挑战.针对行人的姿态变化问题,本文首先对训练集中行人图片进行稠密图像块采样获得图像块集合,然后对每一个图像块提取其局部表观空间特征,最后在此特征集上聚类得到通用的行人部件字典.由于该部件字典编码了行人的部件信息,因此通过该字典内的每一个码元可以建立两幅行人图像中特定图像块之间的对应关系.将两幅行人图片的图像块集合分别向部件字典投影,可以获得2幅行人图片姿态对齐后的图像块序列.针对监控场景的亮度变化问题,本文在姿态对齐后的图像块上分别提取4种颜色描述子,并将不同颜色描述子下的图像块相似性进行分数级组合以获得更好的亮度不变性.其中不同颜色描述子之间的组合系数通过结构化输出支持向量机学习得到.在常用的视点不变行人重识别(viewpoint invariant pedestrian recognition,VIPe R)数据集上的实验结果表明,该方法在存在行人姿态变化和场景亮度变化干扰时获得了较好的行人重识别效果.     

基于局部搜索的多椭圆随机检测算法

随机采样2个图像点后再分别在这2点相关的3条线段上搜索1个图像点。利用采样的2个图像点和搜索的3个图像点决定侯选椭圆。当采样和搜索图像点时,通过识别和剔除孤立、半连续噪声和不在同一个真椭圆上的图像点,显著地减少了无效采样及无效计算。在确认真椭圆时,文中方法将椭圆变换成对应圆,通过确认真圆来确认真椭圆并直接控制椭圆的检测精度。数值实验结果表明:该文算法在检测多个椭圆时的检测效率和鲁棒性等方面都具有良好的性能。     

基于人脸图像的性别识别建模与仿真研究

研究性别识别问题,人脸图像受到光照、姿态、年龄的变化等影响,采用单一特征提取方法难获得较高的性别正确率。为提高性别识别正确率,提出采用几何特征和主成分分析结合的性别识别算法。首先采用几何特征方法对人脸图像的特征进行提取,然后采用主成分分析选择对识别结果有重要影响的特征,最后将选择特征输入到支持向量机进行学习,建立性别分类器。采用印度人脸库对算法性能进行检验,结果表明,本文算法加快了性别识别的速度,提高了识别正确率,能够对光照和姿态变化较大的图像进行正确识别。     

人耳人脸特征融合在身份鉴别中的研究

针对单一人耳识别对姿态变化鲁棒性较差的问题,鉴于人脸在图像性质和生理位置上与人耳具有相似性和互补性,使用了多模态特征融合的方法提高姿态变化下的识别率.与传统的独立成分分析首先获得独立的基向量(ICAl)不同,提出了利用ICA直接获得独立的鉴剐特征的方法(ICA2).在USTB图像库上分别将两种ICA特征进行单模态和多模态的融合.实验表明,两种特征的融合提高了单一模态的识别率,并且多模态识别优于单一的人耳或人脸识别.     

基于图像矩的驾驶员面部椭圆拟合定位

基于人脸类似椭圆的特性,提出一种基于图像矩的驾驶员面部椭圆拟合定位方法.该方法利用肤色信息对驾驶室采集的彩色图像在颜色空间进行快速分割的基础上,利用图像矩拟合驾驶员面部椭圆,从而准确的定位跟踪驾驶员的位置,有效的降低了驾驶室的复杂背景、驾驶员各种面部姿态、表情对人脸检测定位的影响,同时也避免了常用椭圆拟合方法计算量大的不足.实验与仿真表明该方法具有快速、稳健和姿态不敏感的特性.     

联合姿态先验的人体精确解析双分支网络模型

人体解析旨在将人体图像分割成多个具有细粒度语义的部件区域,进行形成对人体图像的语义理解.然而由于人体姿态的复杂性,现有的人体解析算法容易对人体四肢部件形成误判,且对于小目标区域的分割不够精确.针对上述问题,本文联合人体姿态估计信息,提出了一种人体精确解析的双分支网络模型.该模型首先使用基干网络表征人体图像,将人体姿态估计模型预测到的姿态先验作为基干网络的注意力信息,进而形成人体结构先验驱动的多尺度特征表达,并将提取的特征分别输入至全卷积网络解析分支与检测解析分支.全卷积网络解析分支获得全局分割结果,检测解析分支更关注小尺度目标的检测与分割,融合两个分支的预测信息可获得更为精确的分割结果.实验结果验证了本文算法的有效性,在当前主流的人体解析数据集LIP和ATR上,本文方法的mIoU评测指标分别为52.19%和68.29%,有效提升了解析精度,在人体四肢部件以及小目标部件区域获得了更为准确的分割结果.     

基于深度图像的多学习者姿态识别

在数字化学习场景中,人体姿态的识别有助于分析学习者的学习状态。提出了一种基于深度图像的多学习者姿态识别方法。首先通过Kinect的红外传感器获取包含深度信息的图像,利用深度图像进行人像-背景分离;然后提取人体的轮廓特征Hu矩;最后采用SVM分类器对轮廓特征进行分类和识别。实验结果表明,本方法能有效地识别多个学习者的举手、正坐和低头等姿态。     

Projectively invariant decomposition and recognition of planar shapes

An algorithm is presented for computing a decomposition of planar shapes into convex subparts represented. by ellipses. The method is invariant to projective transformations of the shape, and thus the conic primitives can be used for matching and definition of invariants in the same way as points and lines. The method works for arbitrary planar shapes admitting at least four distinct tangents and it is based on finding ellipses with four points of contact to the given shape. The cross ratio computed from the four points on the ellipse can then be used as a projectively invariant index. It is demonstrated that a given shape has a unique parameter-free decomposition into a finite set of ellipses with unit cross ratio. For a given shape, each pair of ellipses can be used to compute two independent projective invariants. The set of invariants computed for each ellipse pair can be used as indexes to a hash table from which model hypothesis can be generated Examples of shape decomposition and recognition are given for synthetic shapes and shapes extracted from grey level images of real objects using edge detection.     

Application of the Fisher-Rao Metric to Ellipse Detection

The parameter space for the ellipses in a two dimensional image is a five dimensional manifold, where each point of the manifold corresponds to an ellipse in the image. The parameter space becomes a Riemannian manifold under a Fisher-Rao metric, which is derived from a Gaussian model for the blurring of ellipses in the image. Two points in the parameter space are close together under the Fisher-Rao metric if the corresponding ellipses are close together in the image. The Fisher-Rao metric is accurately approximated by a simpler metric under the assumption that the blurring is small compared with the sizes of the ellipses under consideration. It is shown that the parameter space for the ellipses in the image has a finite volume under the approximation to the Fisher-Rao metric. As a consequence the parameter space can be replaced, for the purpose of ellipse detection, by a finite set of points sampled from it. An efficient algorithm for sampling the parameter space is described. The algorithm uses the fact that the approximating metric is flat, and therefore locally Euclidean, on each three dimensional family of ellipses with a fixed orientation and a fixed eccentricity. Once the sample points have been obtained, ellipses are detected in a given image by checking each sample point in turn to see if the corresponding ellipse is supported by the nearby image pixel values. The resulting algorithm for ellipse detection is implemented. A multiresolution version of the algorithm is also implemented. The experimental results suggest that ellipses can be reliably detected in a given low resolution image and that the number of false detections can be reduced using the multiresolution algorithm.     

Applying the Hough transform pseudo-linearity property to improve computing speed

This work describes a general method of acceleration of the convergence of the Hough transform based, on the one hand, on an improvement of the image analysis speed, and, on the other hand, on the space undersampling of the image. This method is used in image processing to extract lines, circles, ellipses or arbitrary shapes. The results presented are applied to the detection of straight-line segments and ellipses, but can be extended to any type of transform.     

基于频率筛分的无监督人体姿态特征提取与识别研究

基于视频序列的人体行为分析需要检测和判别人体姿态,已有人体姿态检测与判别方法往往达不到实用性要求.从两个方面探讨应用BEMD(bidimensional empirical mode decomposition)算法提升特征分离度与判别性,以进行人体姿态检测和判别:BEMD分解源图像得到的多层固有模态图BIMF具有判别特征,可形成具有强边缘的对比度高的区域,其中包括人体轮廓区域;从低分辨率尺度BIMF图像到高分辨率尺度BIMF图像递归计算,建立基于BEMD的多尺度树(BEMD muhiscale-trees tructured)模型,快速提取目标区域并获取人体形状轮廓特征.实验证明,利用该方法进行人体姿态轮廓特征提取,并建立人体姿态的简化模型,可快速检测并判别人体姿态,以达到实时识别.     

Effect of head-mounted displays on posture

OBJECTIVE: The aim of the present study was to determine if a wearable system based on a head-mounted display (HMD) causes users to alter their head position and adopt postures that place greater stress on the musculoskeletal system. BACKGROUND: HMDs are common output devices used with wearable computers. HMDs provide the wearer with visual information by projecting computer-generated virtual images in front of the eyes. Deviations of neck posture from a neutral upright position increase the stresses on the musculoskeletal system of the head and neck. METHOD: Seven paramedics simulated the treatment of a patient under a normal condition and when using an HMD wearable computer system. During the simulations a posture analysis was performed using the Rapid Upper Limb Assessment method. RESULTS: The postures adopted when wearing an HMD, as compared with a normal condition, scored significantly higher for the neck (z = 2.463, p < .05) and for overall body posture (left side of the body: z = 2.447, p < .05; right side of the body: z = 2.895, p < .05). CONCLUSION: Wearing an HMD can force the wearers to modify their neck posture. As such, the musculoskeletal system of the head and neck may be placed under increased levels of stress. APPLICATION: Potential users should be made aware that HMDs could dictate modifications in neck posture, which may have detrimental effects and may compound the weight effect of the HMD.     

Multi-ellipses detection on images inspired by collective animal behavior

This paper presents a novel and effective technique for extracting multiple ellipses from an image. The approach employs an evolutionary algorithm to mimic the way animals behave collectively assuming the overall detection process as a multi-modal optimization problem. In the algorithm, searcher agents emulate a group of animals that interact with each other using simple biological rules which are modeled as evolutionary operators. In turn, such operators are applied to each agent considering that the complete group has a memory to store optimal solutions (ellipses) seen so far by applying a competition principle. The detector uses a combination of five edge points as parameters to determine ellipse candidates (possible solutions), while a matching function determines if such ellipse candidates are actually present in the image. Guided by the values of such matching functions, the set of encoded candidate ellipses are evolved through the evolutionary algorithm so that the best candidates can be fitted into the actual ellipses within the image. Just after the optimization process ends, an analysis over the embedded memory is executed in order to find the best obtained solution (the best ellipse) and significant local minima (remaining ellipses). Experimental results over several complex synthetic and natural images have validated the efficiency of the proposed technique regarding accuracy, speed, and robustness.     

一种基于变形模板的椭圆提取算法

针对普通的动态轮廓算法对曲线仅提供有限的连续与正则性约束,虽然带来了局部的灵活性,却丧失了全局的完整性,从而导致轮廓提取结果强烈敏感于图像噪声及邻近边缘点,针对这一问题,给出了一种基于椭圆变形模板的椭圆提取新算法。该算法使用一种基于椭圆形状约束的变形模板,由于其能量最小化过程直接在椭圆的参数空间中进行,从而可保证提取的结果一定是椭圆。此外,算法还容许同时提取多个椭圆,并可充分利用各个椭圆参数及其相关性(例如同心椭圆)等先验知识来实现快速、准确、鲁棒的椭圆提取。仿真结果与实际图像应用表明,该新算法是有效的。     

Relationship analysis between body flexion angles and smartphone tilt during smartphone use

Although smartphones are used as essential devices in everyday life, many users are exposed to joint diseases owing to prolonged use. The objectives of this study were to analyze how posture and smartphone tasks affect various body flexion angles and develop an algorithm to classify posture/task and estimate body flexion angles using smartphone tilt data. Eighteen participants performed two tasks (playing a game and reading news) in two postures (sitting and standing) in a laboratory environment. The three-axis orientation data (azimuth, pitch, and roll) of the smartphone and the participants’ body flexion angles were measured simultaneously. This study found that the cervical, thoracic, lumbar, and overall flexion angles were all statistically significantly different depending on the posture of the smartphone user, and the cervical flexion angle was significantly different depending on the task. Furthermore, task and task × posture can be classified with high accuracy based on smartphone tilt data, and tilt data had a high correlation with body flexion angles. Relevance to industry: The results of this study can be used as a reference for designing various products and interfaces for neck health. The results can be applied as a smartphone alarm or a built-in application, which can inform the user of the need to stretch his or her neck.     

基于体感识别技术的乒乓球握姿训练系统设计

为了提高乒乓球握姿训练的自动化水平,提出一种基于体感识别技术的乒乓球握姿训练系统,采用图像传感器进行乒乓球握姿的图像信息采集,对采集的乒乓球握姿图像信息进行体感特征监测,提取乒乓球握姿图像的体感动态特征量,采用支持向量机模型进行乒乓球握姿的体感特征分类识别。根据对乒乓球握姿的体感特征识别结果进行乒乓球握姿的自动调节,实现乒乓球握姿训练优化。仿真结果表明,采用该方法进行乒乓球握姿训练的提高特征检测能力较好,识别准确性较高,提高了乒乓球握姿的自动训练水平。     

Calculating ellipse overlap areas

We present an approach for finding the overlap area between two ellipses that does not rely on proxy curves. The Gauss-Green formula is used to determine a segment area between two points on an ellipse. Overlap between two ellipses is calculated by combining the areas of appropriate segments and polygons in each ellipse. For four of the ten possible orientations of two ellipses, the method requires numerical determination of transverse intersection points. Approximate intersection points can be determined by solving the two implicit ellipse equations simultaneously. Alternative approaches for finding transverse intersection points are available using tools from algebraic geometry, e.g., based on solving an Eigen-problem that is related to companion matrices of the two implicit ellipse curves. Implementations in C of several algorithm options are analyzed for accuracy, precision and robustness with a range of input ellipses.