车载全景图像快速配准方法研究

发布时间：2018-05-28 01:32

  本文选题：车载全景系统 + 图像配准　； 参考：《重庆理工大学》2017年硕士论文

【摘要】：随着汽车数量的迅速增加,由于驾驶员视觉盲区引发的安全驾驶问题日益突出、交通事故不断攀升。目前,车载全景系统被认为是缓解该问题的有效手段之一,已成为智能交通和图像处理领域近年来的研究热点,从而吸引了众多学者的关注。而图像配准是决定车载全景系统是否可行的关键环节。因此,本文针对车载全景系统中快速图像配准方法开展了深入研究,主要研究内容如下:1、针对传统尺度不变特征变换(SIFT)算法在特征提取与特征描述时计算量大、实时性差的问题,提出一种基于区域分块的SIFT快速配准方法。首先,将匹配图像和待匹配图像分割成若干均匀的子图,通过计算每个子图的信息熵值与设定阈值比较来判断局部子图的特征类型;然后对筛选出来的子图进行SIFT特征提取和对特征向量降维处理,生成PCA-SIFT描述子;最后对两幅图像进行配准和剔除误匹配点对。结果表明,在保证配准精度在90%以上的基础上,配准时间减少了15%-25%。2、针对传统的Harris角点检测算法,手动输入单个阈值可能出现角点聚簇、伪角点等现象。提出了一种自适应阈值的Harris角点检测的图像配准方法,首先将图像分割成3?3个无重叠子图,根据每个子图的对比度的大小,来设置每个子图的阈值。然后采用NCC算法对检测出的角点进行粗匹配;最后采用RANSAC算法将粗匹配中误匹配点对进行剔除。实验表明,该算法使检测的角点分布比较均匀,并在图像配准中有效地增加图像匹配点对数,能有效提高图像配准的准确度。3、为了验证本文的图像配准算法在车载全景系统中应用效果,采用Directshow技术和两个USB摄像头搭载了系统实验平台,分别采用改进的SIFT算法、改进的Harris角点检测算法对采集的图像帧进行提取特征点和图像配准的效果对比。实验结果表明:采用改进的SIFT配准算法,融合后图像效果良好,并能有效减少图像拼接时间。总之,本文研究的图像快速配准算法,可提高图像配准效率20%左右,有效减少了图像配准环节的时间消耗,为研制实用的车载图像全景系统奠定技术基础。
[Abstract]:With the rapid increase of the number of cars, the problem of safe driving caused by drivers' visual blind area is becoming more and more serious, and the traffic accidents are rising. At present, vehicle panoramic system is regarded as one of the effective means to alleviate this problem, and it has become the research hotspot in the field of intelligent transportation and image processing in recent years, which has attracted the attention of many scholars. Image registration is the key to determine the feasibility of vehicle panoramic system. Therefore, in this paper, the fast image registration method in vehicle panoramic system is deeply studied. The main research contents are as follows: 1. For the traditional scale-invariant feature transformation (SIFT) algorithm, the computation of feature extraction and feature description is heavy. A fast SIFT registration method based on regional block is proposed. Firstly, the matching image and the image to be matched are divided into several uniform subgraphs, and the feature types of the local subgraphs are determined by calculating the information entropy value of each subgraph and comparing the information entropy value with the set threshold value. Then SIFT feature extraction and feature vector dimensionality reduction are used to generate the PCA-SIFT descriptor. Finally, the two images are registered and the mismatched point pairs are eliminated. The results show that the registration time is reduced by 15% to 25% on the basis of the registration accuracy is over 90%. For the traditional Harris corner detection algorithm, the single threshold of manual input may appear such phenomena as corner clustering and pseudo corner. An image registration method for adaptive threshold Harris corner detection is proposed. Firstly, the image is divided into 3? 3 non overlapping subgraphs. The threshold of each subgraph is set according to the contrast of each subgraph. Then the NCC algorithm is used to coarse match the detected corner, and the RANSAC algorithm is used to eliminate the mismatched points in rough matching. Experiments show that the algorithm makes the corner distribution more uniform, and effectively increases the logarithm of image matching points in image registration. It can effectively improve the accuracy of image registration. In order to verify the application effect of the image registration algorithm in vehicle panoramic system, Directshow technology and two USB cameras are used to carry the system experimental platform, and the improved SIFT algorithm is adopted respectively. The improved Harris corner detection algorithm is used to extract feature points and image registration. The experimental results show that the improved SIFT registration algorithm is effective and can effectively reduce the time of image stitching. In a word, the fast image registration algorithm studied in this paper can improve the efficiency of image registration by about 20%, effectively reduce the time consumption of image registration, and lay a technical foundation for the development of a practical panoramic image system.
【学位授予单位】：重庆理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U463.6;TP391.41

【参考文献】

中国期刊全文数据库 前10条

1 王丽芳;赵雅楠;秦品乐;高媛;;基于ORB和GroupSAC复杂场景视频图像的快速角点检测[J];科学技术与工程;2017年02期

2 汪军;梁凤梅;;基于P样条和局部互信息的非刚性医学图像配准[J];计算机应用研究;2017年08期

3 李玉峰;李广泽;谷绍湖;龙科慧;;基于区域分块与尺度不变特征变换的图像拼接算法[J];光学精密工程;2016年05期

4 赵小强;岳宗达;;一种面向图像拼接的快速匹配算法[J];南京理工大学学报;2016年02期

5 李钦;游雄;李科;张彦喜;;PCA-SIFT特征匹配算法研究[J];测绘工程;2016年04期

6 张东;余朝刚;;基于特征点的图像拼接方法[J];计算机系统应用;2016年03期

7 张勇;王志锋;马文;;基于改进SIFT特征点匹配的图像拼接算法研究[J];微电子学与计算机;2016年03期

8 杜往泽;宋执环;闫文博;吴乐刚;;单摄像头旋转监控下的快速图像拼接[J];中国图象图形学报;2016年02期

9 刘坤;吕晓琪;谷宇;于荷峰;任国印;张明;;快速数字影像重建的2维/3维医学图像配准[J];中国图象图形学报;2016年01期

10 吴一全;王凯;;基于SUSAN算子和角点判别因子的目标边缘检测[J];中国科学院大学学报;2016年01期

中国硕士学位论文全文数据库 前4条

1 胡玉;车载图像系统中的全景影像的校正与拼接[D];吉林大学;2016年

2 仲明;基于特征点精确配准的图像拼接技术的研究[D];华东师范大学;2015年

3 王则浪;基于DM8148的车载全景拼接算法及视频记录软件设计与实现[D];浙江工业大学;2014年

4 林蔚;基于互信息和蚁群算法的多分辨率二维—三维医学图像配准的研究[D];广西大学;2013年

，

本文编号：1944688