嵌入式系统中图像融合技术研究

发布时间：2018-02-09 05:05

本文关键词： 嵌入式图像配准图像融合 FPGA DSP　出处：《中原工学院》2017年硕士论文　论文类型：学位论文

【摘要】：当代社会科技的迅猛发展迫切需要能够实时获取图像信息并进行综合处理,来满足现代人类高节奏的生活,嵌入式图像处理系统应运而生。图像融合技术是将相关场景的多幅图像合并成为一幅,弥补各传感器成像不足之处获得较大信息量。针对现有图像融合算法仅仅将特征考虑在内的现状,本文提出了结合特征点相似性与空间结构的优化图像融合算法,结合嵌入式发展情况,设计了基于现场可编程门阵列和数字信号处理器架构的嵌入式图像融合系统,说明了系统硬件平台设计和软件算法实现过程。本文结合图像配准和图像融合技术的发展和传感器采集图像的特点,对基于特征点的图像融合算法进行深入研究,提出了以加速鲁棒性特征(SURF)算法为理论依据的改进图像融合算法。基于特征点的图像融合算法首先获取相关场景多幅图像的SURF特征进行特征描述,提取出SURF特征描述子;其次利用SSD(Sum of Squared Differences)对提取的特征描述子进行特征匹配,并根据MSAC(M-estimator SAmple Consensus)去除匹配对中的“外点”求取转换参数;最后利用重叠区线性过渡用来进行图像融合。而基于SURF特征点的改进图像融合算法考虑了图像的空间结构特征,首先根据基于SURF特征点的图像融合算法求取的转换参数,分析求解原数据空间结构,结合特征相似性与空间结构求取复合特征;其次采用ICP(Iterative Closest Point)算法求取匹配矩阵;最后利用双向匹配限制求取最终的匹配矩阵。仿真实验表明,所提方法准确度高,鲁棒性强。针对图像融合在嵌入式平台的使用,考虑其硬件资源,搭建了以现场可编程门阵列和数字信号处理器为基础的硬件平台,设计了以前端视频信号采集子系统、图像融合子系统、图像融合调试子系统为核心的嵌入式图像处理平台,对核心器件选型、PCB设计进行详细描述,在嵌入式系统平台上完成了图像配准和图像融合。
[Abstract]:The rapid development of modern social science and technology urgently needs to be able to obtain real-time image information and comprehensive processing to meet the high pace of modern human life. The embedded image processing system emerges as the times require. Image fusion technology is to combine multiple images of related scenes into one image. Aiming at the fact that the existing image fusion algorithms only consider the features, this paper proposes an optimized image fusion algorithm which combines the similarity of feature points with the spatial structure. According to the development of embedded system, an embedded image fusion system based on field programmable gate array and digital signal processor is designed. The design of hardware platform and the realization of software algorithm are explained. Combining the development of image registration and image fusion technology and the characteristics of image acquisition by sensor, the image fusion algorithm based on feature points is studied deeply in this paper. An improved image fusion algorithm based on the accelerated robust feature surf algorithm is proposed. Firstly, the feature point based image fusion algorithm acquires the SURF features of multiple scene images for feature description and extracts the SURF feature descriptor. Secondly, the extracted feature descriptors are matched by SSD(Sum of Squared differences, and the "outer points" in the matching pairs are removed according to the MSAC(M-estimator SAmple Consensus. Finally, the linear transition of overlapping region is used for image fusion. The improved image fusion algorithm based on SURF feature points takes into account the spatial structure features of the image. Firstly, the conversion parameters are obtained according to the image fusion algorithm based on SURF feature points. The original data spatial structure is analyzed and solved, combining the feature similarity with the spatial structure to obtain the composite feature; secondly, the matching matrix is obtained by using ICP(Iterative Closest Point algorithm; finally, the final matching matrix is obtained by using the bidirectional matching restriction. The simulation results show that, The proposed method has high accuracy and robustness. Considering the hardware resources of image fusion in embedded platform, a hardware platform based on field programmable gate array and digital signal processor is built. The embedded image processing platform based on video signal acquisition subsystem, image fusion subsystem and image fusion debugging subsystem is designed. The PCB design of core device selection is described in detail. Image registration and image fusion are completed on the embedded system platform.
【学位授予单位】：中原工学院
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】