【作者】 徐艳；

【导师】 宗晓萍；

【作者基本信息】 河北大学 ， 检测技术与自动化装置， 2007， 硕士

【摘要】 随着计算机技术、电子技术、图像处理、模式识别的迅速发展,机器视觉的实际研究与应用日益得到重视,并在许多领域取得了骄人的成果,这就增加了机器视觉技术应用于机器人的可能性。本文是以机器人视觉伺服为最终研究目标,利用神经网络技术和图像处理技术,完成了机器视觉研究中的三项内容：1)根据神经网络和摄像机标定的特点,使用BP神经网络完成不需要预先建立成像系统的结构模型的平而摄像机标定,它利用网络的自学习和自适应能力,通过与外界信息的交互作用,形成一种非线性映射,以表示图像和空间坐标之间的映射关系,由于其快速的并行计算能力及良好的泛化和局部逼近能力,此方法具有了较强的标定实时性；2)在图像处理部分,描述了融合像素点的梯度、相角和方向特征多种信息,利用简单的比较、加减等模糊逻辑运算来完成含有强噪声的图像的边缘检测,这种方法尤其适合对实际工作中含有强噪声的图像进行边缘检测；3)最后,使用7个不变矩作为模式识别的特征值,利用BP网络取代原始的统计分类器,简单、快速、有效地完成物体的模式识别工作。本文利用简单物体进行了模式识别实验,验证了研究工作的正确性。神经网络的采用理论上可以有效解决视觉伺服系统中速度问题这一瓶颈；图像处理中各种抑制噪声的手段可以提高系统的精度,有利于整个机器人系统速度的提高。更多还原

【Abstract】 With the rapid development of Computer Science, electronic technology, Image Process、Pattern Recognition, people are paying more and more attention on the robot vision’s research and application which has resulted in excellent results in many field, and this cause more probability that people can use the machine vision on robot.In this paper we study the key technology for robot vision system based on the robot vision servoing system model. Using neural network and the image process technique, three parts were completed:1) Based on the characters of neural network and camera celebration, using BP network to complete the camera celebration which doesn’t need to build the structure model of camera system. Using the capability of self-study and generalization of neutral network to build the relationship between the image point and the dimensional point. The technology of camera calibration based on the neural network can cause the accuracy of the celebration and the speed is higher than former.2) In the part of image processing, an algorithm which can detect the edge in the high noisy image was improved. In this algorithm, several kinds of information are used as the characteristics of edge detection, and fuzzy logic is used to fuse the information together. So the edge detecting process is more effective and gives much better edge information especially for the high noisy image.3) Finally we selected seven features for pattern recognition using the moment invariant features, and then these features are used as the input of the BP network to complete the recognition.The application of neural network can speed up the process of robot vision servoing system effectively and the method on how to reduce the noise in the image can improve the precision of the system.更多还原