导　　师： 杨汝清

学科专业： 080202

授予学位： 博士

作　　者： ;

机构地区： 上海交通大学

摘　　要： 本文对220kV//330kV高压带电清扫机器人视觉系统中的若干问题进行了研究。高压带电清扫机器人是用于220kv//330kv高压变电站环境中,对高压绝缘瓷瓶的污秽进行带电清扫的自动化设备。为了实现清扫操作的自动化,这里采用机器人视觉系统解决绝缘瓷瓶的识别和定位问题。主要研究工作如下: 给出了机器人视觉系统的两种解决方案通过对绝缘瓷瓶带电清扫操作及其工作环境的分析,给出了两种机器人视觉系统解决方案:手-眼机器人视觉系统和远距离机器人视觉系统,详细介绍了两种方案的具体实现方法。从以上方案中提取出若干问题作为本文的主要研究内容。这些问题包括:1.绝缘瓷瓶的特征表示和特征不变性问题;2.绝缘瓷瓶的识别和定位问题;3.人工标志物体的视觉跟踪问题。 提出一种基于SUSAN原理的尺度不变性特征/(以下简称SESIF特征/)绝缘瓷瓶是由光滑曲面组成的、釉质表面的弱纹理物体。绝缘瓷瓶位于室外变电站环境中,其图像会受到环境光照变化、摄像机内参数变化、视角变化、杂乱背景、噪音、部分遮挡、表面材质和纹理等因素的影响。本文应用SUSAN原理、尺度空间理论和特征描述子方法,提出一种基于SUSAN边缘的尺度不变性特征/(SESIF特征/),用于绝缘瓷瓶的特征表示。文中给出了SESIF特征的探测算法,通过实验系统评估了该特征的性能,与其它不变性特征作了对比实验。这些不变性包括:视角不变性、尺度和旋转不变性、图像模糊不变性、噪音不变性、图像压缩不变性、线性和非线性光照不变性。实验证明了该特征的有效性。 应用SESIF特征解决了绝缘瓷瓶的识别和定位问题本文首先研究了基于SESIF特征的匹配和约束问题,包括: 1.引入SNN算法生成匹配假设,通过实验证明:在合适的阈值的条件下,SNN算法的计算效率要优于传统的k-d� This dissertation presents a research on several important problems of robot vision system of a 220kV//330kV high-voltage live-line cleaning robot /(HLCR/). HLCR is a semi-automatic system to clean insulators in outdoor 220kV//330kV high-voltage transformer substation. In order to implement automatic operation, a robot vision system will be applied to estimate the related posture between insulator and robot. The main research is as follows: Presenting two solutions on robot vision system Based on an analysis of HLCR, two solutions of robot vision system were given: eyes-on-hand robot vision system and remote robot vision system. Several problems coming from above solutions had been studied, which included: 1.How to represent invariant features of insulators? 2. How to recognize and localize insulators? 3. How to estimate the motion of a 3D rigid marker? Presenting a SUSAN edge-based scale invariant feature/(SESIF/) Being the weak textured object with curved smooth surface and locating in outdoor environment, a representation of insulator must consider some factors, for example: changed on illumination, camera parameters, viewpoint, cluttered background, material and texture of object. In this dissertation, we presented a SUSAN edge-based scale invariant feature /(SESIF/) to describe insulators, which was derived from SUSAN principle, discrete scale-space theory and local feature descriptor. The algorithm of SESIF was given and a series of invariant experiments of invariant including comparison with other invariant features on viewpoint, scale and rotation, image blur, noise, JPEG compression and illumination was done too. Recognizing and localizing weak textured object using SESIF Firstly, matching and constraints on SESIF were approached, as follows: 1.SNN algorithm was used to find matches, but k-d tree. Our experiments proved that SNN is less expense than k-d tree with a suitable boundary condition; 2.A peak number constraint was presented to eliminate incorrect matches; 3.General Hough transfor

关 键 词： 机器人视觉 尺度不变性 特征 物体识别 视觉跟踪

领　　域： [自动化与计算机技术] [自动化与计算机技术]