基于拓扑图及三维特征的三维物体识别方法

提出一种基于拓扑图及三维特征的三维物体识别方法.该方法在图象预处理阶段先检测物体可视表面的闭合边界,再依据闭合边界对物体进行表面分割,并用一个景物特征图对景物进行表达(SAGR).在建模阶段,先采用多视图表面模型方法建模,再依据表面分割,对模型用一个模型特征图进行表达(MAGR).在物体识别阶段,将景物与模型的匹配分3部分进行,即扫描部件、图的匹配部件及匹配检验部件.匹配首先是基于 SAGR 与 MAGR 图的拓扑性质的匹配,然后是基于三维特征的匹配.根据匹配结果,或对未知物体自动建模,或给出识别结果.     

基于TMS320DSP的装配视觉检测系统

讨论了基于ＴＭＳ３２０２０ＤＳＰ的装配视觉检测系统的组成、工件识别方案及工件图象的处理方法。     

计算机视觉摄像机定标中投影矩阵的计算

摄像机定标是计算机视觉的基本问题，对于计算机视觉摄像机定标中投影矩阵的计算进行了总结和证明。     

基于多层前馈型神经元网络模型的摄像机定标算法

针对传统线性、非线性摄像机定标算法的鲁棒性,提出了基于多层前馈型神经元（MLFN）网络模型来替代传统精确的摄像机定标数学模型,MLFN网络模型作为层状网络可实现任意维复杂的输入／输出映射,对于无需计算摄像机构,外参数的应用场合,该模型提供了一种实用且有较好鲁棒性的摄像机定标算法,同时为了补偿摄影机非线性畸变,把图像按畸变程度分割成两个区域,分别建立各自基于MLFN网络的摄像机定标模型,实验表明,该方法有效补了畸变,并提高了模型精度,给出了基于MLFN网络模型摄像机定标算法实验,验证了该模型的有效性。     

Multi-objective dynamic detection of seeds based on machine vision

An approach to inspecting massive numbers of moving seeds was studied based on the techniques of dynamic inspection and machine vision. A progressive scanning CCD camera with external trigger function was used for real-time capture of dynamic images of seeds. The methods based on R channel of RGB (Red, Green and Blue) and region-dependent segmentation were adopted to reduce the data size of image processing and improve the efficiency of seeds inspection. All the seeds were sorted into four grades according to their morphological characteristics, such as surface area, perimeter, major axis, minor axis, circularity and eccentricity. The detection experiments indicated that the eligible ratio of the classifications was about 81.90% by this real-time inspection system.     

SAR图像的特征提取与目标识别

针对已有的SAR图像自动目标识别方法效果不太好的问题，提出了SAR图像的自动目标识别三种方法，即采用灰度特征的目标识别，采用栅格特征的目标识别，和采用模版特征的目标识别.理论和实验表明，采用模版特征的目标识别效果比较好.     

小波变换在多仿生机器鱼视觉系统中的应用研究

针对集控式多仿生机器鱼系统（multiple robot fishes coordinate System，MRFS）的实际要求和特点，采用小波变换对图像进行去噪处理．利用图像与噪声的小波模极大值随尺度变化大小不同，通过极大模点在各个尺度的取值来确定噪声的位置和特征，从而完成去噪目的．     

热超声倒装键合机视觉系统的设计与实现

针对热超声倒装键合过程中对准工作的特点,确定合理的视觉自动对准方案,在对多种颜色和形状的LED光源进行实验对比的基础上,设计环形LED光源与单个LED发光二极管组合的照明方式,使芯片、基板表面亮度一致且表面几何特征清晰。视觉软件以HexSight为核心,利用创建参考点实现不同目标的识别,并对2个CCD摄像机的控制方案进行优化。通过分析多次识别结果对单项误差进行评估,并以实验用1 mm×1 mm的表面有8个凸点的芯片为例计算视觉识别系统的综合误差,结果表明此视觉识别系统满足芯片键合时对定位精度的要求。     

Neural correlates of short-term perceptual learning in orientation discrimination indexed by event-related potentials

The current work investigated the neural correlates of visual perceptual learning in grating orientation discrimination by recording event-related potentials (ERPs) from human adults. Subjects were trained with a discrimination task of grating orientation in three consecutive training sessions within 2 h. While reaction times (RTs) were shortened gradually across training sessions, the N1 was decreased and the P2 was increased over the parietal and occipital areas. A broadly distributed P3 was increased along with more practices. In addition, the time course of learning reflected in the P2 and P3 amplitudes was in line with the changes of reaction times and exhibited a stable level during later training. The impli- cations of these results to the neural mechanisms subserving perceptual learning were discussed.     

机器视觉系统的设计

介绍了机器视觉系统的概念,提出了机器视觉系统的典型组成,在进行用户需求分析的基础上,详细讨论了如何利用光源、光学镜头、摄像机、图像采集卡以及图像处理软件等主要部件来设计开发一个焊缝机器视觉应用系统的关键技术.最后给出了一个焊缝检测应用实例.