立体视觉中误匹配滤波方法的研究

从二维空间和三维空间2种角度研究误匹配滤波算法,提出在匹配前用于降低误匹配的灰度预处理算法和一种基于真实控制点的视差滤波算法。前者只针对2幅图像的重叠区域进行灰度均衡,可以减少计算量,后者在传统视差均值滤波的基础上可进一步提高误匹配的滤波效率。基于真实图像的实验结果表明,新算法可以有效滤除误匹配,提高三维重建精度,保证重建效果。

基于外极线分区的动态立体匹配算法

基于区域的立体匹配算法仅针对支撑窗内的灰度信息定义匹配代价函数,导致在弱(无)纹理区域采用WTA优化出现歧义性。该文在外极线分区的基础上,改用区域作为匹配基元,针对歧义性区域,在代价函数中引入遮挡项和平滑项,并按照区域优先级的高低,动态匹配相应区域,获得可靠的视差信息。实验证明,该算法在保持实时性的同时对弱纹理区域处理具有有效性。

基于立体视觉的移动机器人楼梯识别

本文提出了一种结构化环境下,基于立体视觉的机器人楼梯识别算法,并将算法该应到自主移动机器人上。该算法首先利用二维图像分析的方法搜索楼梯的疑似区域;进而利用立体视觉对各个疑似区域进行精确三维重建,结合三维信息重构楼梯平面,排除虚假疑似楼梯区域;最后判定机器人和楼梯的相对位姿关系,引导机器人爬楼梯。最终我们将该算法应用到了自主移动机器人上,通过在各种光照条件下的实验,进一步验证了该算法的准确性和快速性。

基于视觉伺服的输电线巡检机器人抓线控制

巡检机器人在越障时,需要完成机器人手臂的准确抓线控制.结合输电线的几何特征和摄像机成像原理,提出了一种基于单摄像机的立体视觉方法来确定输电线的位置和姿态.基于该定位方法及视觉伺服理论,建立机械手抓线伺服控制模型.利用自行研制的巡检机器人进行了视觉伺服抓线实验;实验结果验证了该方法的有效性.

基于立体视觉的虚拟机械臂平面定位研究

结合星球探测的应用背景,对漫游车的工作方式进行了研究,针对车载机械臂开发了一套基于立体视觉的机械臂平面定位仿真系统。该系统依靠虚拟现实技术,通过虚拟机械臂对三维重建得到的平整物体表面的定位仿真得到机械臂的各关节参数,以此指导真实机械臂的运动。论述了基于立体视觉的机械臂定位机理和基于OpenGL的虚拟机械臂的实现过程。采用VC++构建了仿真实验平台,进行了定位实验,获得了较高的定位精度。

基于模型的位姿估计中优化方法研究

介绍了基于模型的位姿估计中所使用的一些优化方法。为了提高位姿估计的精度,摄像机的标定参数必须足够精确,这就对标定过程的非线性优化算法提出了很高的要求,采用了一种新的优化目标函数,用来最小化控制点间的三维重建误差,从而使标定参数是全局最优;在双像机位姿估计中,引入了实时遗传算法进行全局搜索,加快了算法的收敛速度。最后的实验证明了这些方法的正确性并显示出这些方法在精度上比传统方法有了较大程度的提高。

基于立体视觉的移动机器人自主导航定位系统

描述了一种基于立体视觉的移动机器人自主导航定位系统。该系统采用双目立体视觉完成环境特征的 3D信息提取 ,实时计算出机器人相对作业目标的位姿 (6D)关系 ,导引移动机器人控制系统按目标导向进行运动。系统在相对位姿计算中采用旋动(Screw)理论 ,将带约束的多变量函数的非线性优化问题转化为线性方程组的最小二乘问题 ,简化了计算复杂性。实验表明 ,这个导航定位系统在定位精度和数据处理速度上均可满足机器人导航的要求。

虚拟现实辅助机器人遥操作技术研究

本文以水下机器人的遥操作作业为应用背景 ,提出并实现了虚拟现实技术和视觉感知信息辅助机器人遥操作实验系统 .该系统使用了 CAD模型和立体视觉信息完成遥操作机器人及其作业环境的几何建模和运动学建模 ,实现了虚拟作业环境的生成和实时动态图形显示 .采用了基于立体视觉的虚拟环境与真实环境的一致性校正、图形图像叠加、作业体与环境位姿关系建立、基于网络的监控通讯等关键技术 .在这个实验系统中 ,操作人员可利用所生成的虚拟环境 ,在多视点、多窗口作业状态图形和图像显示帮助下 ,实时动态地进行作业观测与机器人遥操作与运动规划 ,为先进遥操作机器人系统的实现提供了经验和关键技术 .

面向月面探测作业的双目立体视觉测量及三维环境生成技术研究

星球漫游车技术是一项跨学科的综合性技术，它反映了一个国家整体科技水平和高技术产业的发展水平，对漫游车进行研究具有长远的战略意义。漫游车一般都工作在特殊的环境中，综合考虑功耗和有效载荷等问题，可以发现，非视觉传感器与视觉传感器相比在环境感知能力和综合性能方面都不甚理想，因此，基于视觉传感器的立体视觉定位也一直是各类漫游车用以感知环境所首选的技术。基于立体视觉可以构建相应的地面人机交互系统。该交互系统具有实现离线示教功能，通过虚拟现实和立体视觉技术实现虚拟漫游车在虚拟地形中行走，虚拟机械臂对虚拟物体进行研磨等功能。操作人员在虚拟环境中对月球车的各种运动进行验证以后，再将数据上传给真实的漫游车以实现真实的漫游和考察工作。因此，立体视觉系统性能的好坏将直接影响到漫游车的后续工作情况，它是漫游车自主导航和进行科学考察的基础，直接关系到漫游车的安全，是一项关键的技术。 本文的研究依托于中国科学院知识创新工程方向性项目“月球车月面探测作业支撑技术研究”，旨在探索立体视觉系统的实现及其与车载机械臂配合工作的问题。从理论和实践两个方面，对其中的若干关键技术，如双摄像机标定、对极几何恢复、图像校正、图像匹配、三维重建和车载机械臂视觉定位等进行了研究。主要研究成果如下： 1．提出一种基于坐标排序的标定特征点存储算法，使得两幅图像中的标定特征点能够快速准确的对应存储。基于径向一致约束（RAC）模型优化方法与直接光学法的优点，提出一种改进的光心确定方法，实验表明该方法有效地提高了Tsai两步法的标定与测量精度。以上两种方法为实现双摄像机的高精度标定奠定了基础。 2．在双摄像机的标定阶段，提出一种新的以三维重投影误差为优化目标的双摄像机两步标定算法，实验证明该方法的标定与测量精度要高于传统的基于二维重投影误差的方法。提出一种考虑多个畸变系数基于三维重投影误差的两步标定算法，实验证明该方法比只考虑一个畸变系数的Tsai两步法能够获得更高的标定与测量精度。研究了基于遗传算法和粒子群算法的双摄像机内、外参数的整体优化问题，实验证明可以获得比传统梯度下降法更高的标定与测量精度。以上改进方法可以使标定精度达到工程要求，并为后续的图像校正和三维测量奠定基础。 3．从简单背景和复杂背景两种情况对对极几何的离线恢复问题进行了研究。对于对极几何的离线恢复问题，提出了基于正确匹配标定特征点进行基础矩阵高精度估计的算法。该算法的特点是与标定共用同一组数据，其有效性得到了实验的验证。对于对极几何的在线恢复问题，基于摄像机成像几何原理证明了图像坐标系下存在的一种左右图像坐标间的约束关系，并设计了相应的稀疏匹配算法，采用两阶段鲁棒稀疏匹配算法对基础矩阵进行了高精度的估计，实现了对极几何的在线恢复。与传统方法相比，本文的方法速度更快，精度更高。 4．研究了标定与非标定图像的极线校正算法，利用上述改进方法求得的高精度的投影矩阵和基础矩阵，获得了令人满意的校正效果，为稠密匹配奠定了坚实的基础。研究并实现了四种稠密匹配算法，并进行了算法性能的比较。针对稠密匹配后的初始候选匹配点对，研究了三种误匹配的剔除方法。实验结果表明这些方法能够有效剔除误匹配点对，提高三维重建的效果。 5．研究了基于最小二乘法和立体成像几何法的三维重建算法，研究了一种集成上述多种改进算法的欧式三维重建算法。研究开发了三维重建软件，基于真实图像进行了欧式三维重建实验研究，取得了良好的重建效果。 6．以月球车为应用对象，研究了虚拟机械臂和车载机械臂的平面定位机理，构建了基于立体视觉的虚拟机械臂定位仿真平台。基于虚拟环境下验证好的数据，进行了月球车车载机械臂平面定位的实验，结果证明车载机械臂具有良好的视觉定位精度。

基于立体视觉的室外非结构化环境障碍物检测方法研究

随着智能机器人系统的发展，机器人的在线感知能力日益受到重视。障碍物检测能力是机器人在线感知能力的一个重要组成部分。因视觉传感器具有独特优势，基于视觉的障碍物检测方法成为目前关注的重点。 室外非结构化环境因结构复杂，机器人缺乏可有效利用的先验知识描述，导致众多障碍物检测系统在该环境中不能有效工作。本文采用全局-局部策略对场景进行由粗到精的分析，弥补室外非结构化环境先验知识不足的难题，提高机器人的在线感知能力。根据该策略，本文在基于视差图的障碍物检测系统框架中，引入视差投影图模块，提出了基于视差投影图的障碍物检测系统框架。该框架在视差投影图模块中全局分析场景视差分布水平，在立体匹配模块中局部分析场景前景目标的几何轮廓信息。依据该框架，针对实际应用中遇到的各种问题，提出了工作于室外非结构化环境的障碍物检测算法。该算法具有如下特点： 1、通过分析视差投影图的地面关联线信息，获得场景的视差分布水平。该信息一方面用来动态更改匹配算法的视差搜索范围，增强算法的实时性和鲁棒性；另一方面用来移除背景地表，简化障碍物分割过程； 2、采用双域滤波抑制噪声，获得清晰的边缘特征，降低立体匹配算法在深度不连续性区域的匹配难度； 3、借助逆向重投影的思想重采样扫描图像，在立体匹配前等效地实现了立体匹配过程中动态变更视差搜索范围的操作； 4、采用基于连通成分的扩散方法替代传统的SAD局部匹配算法，获得高质量的视差图，最终改善障碍物检测的精确性。 在室外非结构化环境中，本文对该算法进行了实验验证。通过设置不同的基线长度，验证了算法在不同的感知距离内的有效性。经实验证明，本算法在一定距离范围内能够有效的检测出障碍物。

基于模型的立体视觉的优化方法

针对 3D视觉检测中最典型的误差源图像中像素的量化误差 ,在已知模型的条件下 ,对双目立体视觉提出一种模型。利用该优化模型及改进的约束最小二乘法对未知模型的立体视觉解法做一些改进 ,可以明显减小系统的误差。仿真结果表明 ,与未知模型的立体视觉方法相比 ,在同样的量化误差条件下 ,该方法可将系统误差减小 50 %以上。

机器人插件作业的视觉导引方法研究

插件作业 (parts mating)是装配机器人的一项基本作业环节 .本文介绍了以双目立体视觉实现该作业的视觉导引方法 .该方法通过采用人机交互方式 ,借助于人的智慧 ,提高了图像特征提取和匹配的准确性和可靠性、可直观准确地给出插件作业的动作参数 ,克服了自动视觉计算复杂、鲁棒性差的缺点 ,适用于机器人遥操作作业 .实验表明 ,基于人机交互的机器人插件作业在立体视觉导引下是完全可行的

块体理论赤平投影解析法与关键块体滑落概率研究—以锦屏二级水电站皮带机隧道为例

Block theory is an effective method on stability analysis of fractured rigid rock mass. There are a lot of discontinuous planes developed in rock mass of Jinping II hydropower station conveyor tunnel, so the stability of conveyor tunnel is related with whether there are unstable blocks on excavation planes. This paper deals with the stability of conveyor tunnel with stereo-analytical method for block theory on the basis of detailed investigation of rock mass data, and makes judgements on the movable blocks sliding types which are induced by all rock discontinuous planes and every excavation plane of conveyor tunnel. A conclusion is obtained that the sliding type of blocks is mainly single sliding, and a relatively few sliding types of double-sided sliding and vertical block falling; Also, the obvious statistical distribution result on movable blocks in conveyor tunnel indicates that there are a bit more instability blocks in left wall, left and right arches than right wall. In this paper, the stochastic probability model is drawn into block theory to study the sliding probability of key block on the basis of detailed investigation of its rock mass data and the development of the discontinuous planes in rock mass of Jinping II hydropower station conveyor tunnel. And some following conclusions are obtained. The relationship between trace length and the probability of instability of key block is inverse ratio. The probability of 1-3m primary joints are relatively higher. Key block containing joints J2 is relatively stable and the reinforcement of the arch would be crucial in the conveyor tunnel. They are all useful to offer effective reinforcement design and have important engineering values.

烟草依赖者的注意偏向研究

Relapse has been a great challenge in clinical treatment and experimental studies of drug addiction. Recent studies suggest that psychological dependence may play a major role in addiction relapse, even more important than physiological dependence. Then a fundamental question arises: how to measure the psychological dependence? How to examine whether an addict has psychologically quitted when leaving drug rehabilitation centers? Self-report, a commonly used evaluation approach, is inevitably vulnerable to various cognitive influences, particularly in explicit tasks. Therefore, an objective index is necessary to evaluate the subliminal psychological drug dependence level. The objective of the current study was to develop such a psychological paradigm to probe the unaware attentional bias of in smoking addicts. Experiment 1 adapted the interocular suppression technique of binocular rivalry to study the attentional bias to cigarette pictures in smokers and age-matched nonsmoker. Results show that the smokers demonstrated similar attentional bias in both visible and unaware conditions, while non-smokers showed attentional bias only in the visible condition, and there was a significant interaction between experiment conditions and subject groups. These results provide compelling evidence for addiction-specific attentional bias in cigarette smokers, by minimizing the influence of confounding conscious factors. Furthermore, attentional bias of smokers in unawareness state was negatively correlated with their cigarette dependence levels, while their pre-test cigarette craving levels was positively correlated with their attnetional bias in the visible condition. This pair of correlations further demonstrated the advantages of unawareness state in disclosing stable dependence states, therefore supporting the effectiveness of the paradigm used in this study. Another interesting finding of Experiment 1 is that non-smokers also showed attentional bias in the visible condition. To exclude the possibility that the attentional bias found in experiment 1 was task-specific, experiment 2 adapted the most commonly-used visual dot probe task with smoking scenes as in relevant reference. The result in experiment 1 was well replicated, i.e., nonsmokers in experiment 2 also showed significant attentional bias to smoking-related stimuli, We interpenetrate this interesting finding as an effect of environmental influence, as the participants of the current study live in a highly smoking-exposed and smoking-encouraged environment, which is quite different with the participants of studies reported in the literature. A series of questionnaires and scales administered in the current study indeed show that most smokers smoked due to influence of the environment. They also acknowledged that smoking as an important media of social communication in China, and even considered that away from the smoking environment would effectively help them to quit. The current study also found that the disgust level towards cigarette pictures and smoking-related scenes of non-smokers was positively correlated with their attnentional bias in the visible condition of experiment 1. It is likely that in a highly smoking-encouraged environment, the remaining few on-smokers have severe disgust to cigarettes and smoking scenes; and their attentional bias might be caused by disgust avoidance. In conclusion, the current study represents the first study showing the existence of unaware attentional bias to smoking related stimuli in cigarette smokers by applying the interocular suppression paradigm, providing a reference to study of dependence of other drugs. The current study also found that our non-smoking participants also showed attentional bias to smoking related stimuli, which may be due to the possible influence of highly smoking-exposed environment of our participants.

品牌形象的消费行为学分析

Brand image is the attributes set and the related associations of a brand in consumers mind, and it is the subjective reflections of brands. The paper explored the factors of brand image system and their weight. As the traditional means to evaluate weight coefficients are not perfect, a new method, conjoint analysis, was attempted. The factors of brand image were explored through questionnaire. Sports sneaker, toothpaste, and personal stereo were chosen as product sample, and four hundred and twenty university students from Tangshan city and Beijing as subjects(each person evaluated two kinds of products). The first two kinds of products were requisites of students, and sports sneaker belonged to High conspicuous products and toothpaste was low conspicuous product. On the other hand, personal stereo was the sample of development and entertainment products. Several factors of three products brand were taken out with factor analysis. In order to explore the weight of the brand image factors, a contrast of factor contribution ratio method, holistic quartation method and conjoint analysis is made here. Twenty university students evaluated the weight of the image factors of three kinds of brand with holistic quartation method, then they gave the weight of personal stereo with conjoint analysis method. Product function, advertising and propaganda, symbolic meaning, market orientation, brand appetency, consuming experience are the factors of sports sneaker brand image. Product function, advertising and propaganda, market orientation, product grade, corporation image are the factors of toothpaste brand image. Corporation image and product function, advertising and propaganda, consuming experience, symbolic meaning, price and function ratio are the factors of personal stereo brand image. So the hypothesis was proved that brand image is an ordinal and organical system, "ordinal" means the weight of factors are different, "organical" means that brand image can be deposed into several factors and the factors belonged to function components and meaning components (function components are the factors about physical characteristics and function, which are called "hard factors"); meaning factors are those that can show the personality、value and lifestyle of consumers, which are called "soft factors". The research also gave evidence of the hypothesis below: the factor structures of brand image of different product category have commonness and individuality; the function components of low conspicuous products are more important than the high conspicuous products. The exploration of conjoint analysis is what the paper seeks to be some creative in some degree.