【作者】 王旭东；

【导师】 王从庆；

【作者基本信息】 南京航空航天大学 ， 模式识别与智能系统， 2008， 硕士

【摘要】 随着人类科技的不断发展,机器人的应用场合越来越广泛,而抓取操作是必不可少的环节。模仿人类的抓取操作,本文研究的是机器人抓取物体过程中的预抓取阶段,即机器人根据视觉传感器感知被抓取物体与抓取有关的参数,并根据机器人手的姿态,采用何种预抓取模式的决策阶段。首先根据BH-4手和Rutgers手设计出一种具有四个手指的多指手模型,并根据Cutkosky提出的人手抓取分类学,将多指手的预抓取模式分为13类。其次,选择26种实物,每两种实物对应一种预抓取模式。在相同环境中,每种实物在五个不同的角度分别采集一幅图像。经过一系列的图像处理方法,提取被抓取物体的姿态、大小、形状以及表面粗糙度特征作为该物体的特征参数用于预抓取模式分类。随后本文着重研究了聚类分析在多指手预抓取模式分类方面的应用。先用模糊C-均值聚类算法验证聚类分析的可行性,在聚类正确率较低且不稳定的情况下,通过改进算法提高性能。在此基础上,本文提出了一种新的改进算法—二阶段加权模糊C-均值聚类算法,经仿真实验分析,其各方面性能的提高都很明显,特别是聚类正确率稳定在96.15%。接着本文深入研究了核方法在特征选取和聚类分析中的应用。并提出了一种结合核主成分分析和核模糊C-均值聚类的新算法,将该算法用在多指手预抓取模式聚类分析中,算法的实时性和正确率都较为理想。最后,本文利用OpenGL和Visual C++开发了多指手预抓取三维仿真平台,对机器人预抓取的过程进行可视化仿真。更多还原

【Abstract】 With the continuous development of science and technology, the robot applications become more and more extensive, and grasping manipulation is essential to the robot applications. Imitating the grasping manipulation of human, the pre-grasping phase of the robot is researched in the paper, it is a decision process, in which the robot perceives the grasping relevant parameters of the object by the vision sensors, and grasps the object using one of the pre-grasping patterns, according to the robot hand posture.Firstly, the four fingered hand model is designed with reference to the BH-4 and Rutgers robot hands. According to the Cutkosky grasp taxonomy, the multi-fingered hand pre-grasping patterns are divided into 13 categories.Secondly, 26 kinds of objects are selected to the research, and every two objects are corresponding to a pre-grasping pattern. In the same environment, each kind of object is collected an image in five different angles. After a series of image processing, the gesture, size, shape and surface roughness characteristics of the object are extracted as characteristic parameters for the pre-grasping pattern classification.Then the application of the clustering analysis in the multi-fingered hand grasping pattern classification is emphatically studied. The feasibility of the clustering analysis is verified by using the fuzzy c-means clustering algorithm. Because of the low clustering accuracy rate and the unstable clustering results, the improved algorithms are studied. Based on these algorithms, a new improved algorithm– the two-stage weighted fuzzy c-means clustering algorithm is presented, various performance indexes of this algorithm are improved obviously, and its clustering accuracy rate is 96.15%, especially.Then the kernel method in the applications of feature extraction and clustering analysis are further studied. A new algorithm, combined with the kernel principal component analysis and the kernel fuzzy c-means clustering algorithm, is presented. The real-time performance and the clustering accuracy rate of the algorithm are both perfect.Finally, the 3D simulation platform for the pre-grasping of the multi-fingered hand is developed by using OpenGL and Visual C++.Visual simulation of the pre-grasping process of the robot hand is demonstrated in the platform.更多还原