【作者】 张斌；

【导师】 刘桂华； 李杰；

【作者基本信息】 西南科技大学 ， 集成电路工程（专业学位）， 2015， 硕士

【摘要】 图像拼接目的是将有衔接重叠的图像拼成一张全景图像,它是计算机视觉、计算机图形学等多学科综合应用的一门技术。图像拼接技术在虚拟现实、遥感技术、机器人技术等诸多领域广泛应用。本文搭建了基于嵌入式系统的巡航机器人图像拼接系统,实现了Linux操作系统移植,并通过无线网络将巡航机器人采集的视频图像传输到PC端进行图像拼接。针对巡航机器人采回的视频图像存在噪声问题,选择中值滤波进行图像去噪处理并对特征点检测的SUSAN角点检测、SIFT特征点检测、SURF特征提取算法进行了对比研究,实验验证了SUSAN角点检测算法检测特征点的准确性高、特征点检测耗时短。本文还选择了实时性、准确性较好的SIFT算法进行特征点的匹配。为了解决将配准后的图像直接拼接存在拼缝问题,本文研究了基于像素的图像融合方法、基于小波变换的图像融合方法等常用的图像融合算法,通过图像融合的评价标准及信息熵、平均梯度、空间频率、标准差的大小对每种融合算法进行比较,最后基于小波变换融合算法进行图像融合处理,实现了图像的无缝拼接,并在巡航机器人上完成了图像拼接系统的实验验证。更多还原

【Abstract】 Image mosaic is a comprehensive application technology involving computer vision, computer graphics and other multidisciplinary,aiming at stitching overlapping images into a panoramic one.Image stitching technology has widely used in virtual reality, remote sensing technology, robotics and other fields.In this paper,we build a test platform for Patrol robot image stitching,transplant the Linux operating system to the ARM platform, transfer video image turned back by Patrol robot to PC through the wireless network platform and finally finish image splicing processing on PC. On account of the noise problems existing in the video images returned by Patrol robot, this paper uses median filter to de-noise image and have compared the SUSAN corner point detection,SIFT feature point detection and SURF feature extraction algorithm, verifying that SUSAN’s detection accuracy is higher and the time is shorter. Meanwhile,we also realize feature points matching by SIFT algorithm which is real-time and accuracy.In order to solve the joint problems, this paper makes a study on fusion methods based on pixel,wavelet transform and other frequently-used fusion algorithms, then compares those methods by information entropy, average gradient, spatial frequency, the size of the standard deviation and other evaluation criterion, and finally realize image stitching by wavelet transform fusion algorithm,We also finish image mosaicing system experimental verification on the Patrol robot.更多还原