摘要
在分析深海序列图像具有时间和空间上连续性的基础上,提出了一种基于卡尔曼滤波跟踪序列图像的空间位置实施序列图像的快速匹配方法。以序列图像间相似变换矩阵的4个参数为状态向量建立卡尔曼滤波器。首先根据前序序列图像匹配后的状态向量预测当前帧待拼接图像的状态向量,建立相对于前一帧图像的相似变换矩阵;然后建立前后帧图像中子区域的对应关系,分别实现子区域内特征点的局部匹配;最后根据局部匹配的结果估计相似变换矩阵,并修正状态向量。实验中,为提高匹配的可靠性,对每个子区域进行放大,当子区域取128×128,放大系数为1.5时,本文提出的局部匹配方法得到的匹配准确率超过90%,匹配效率提升约50%,表明本文方法能有效地应用于序列图像的拼接。
On the basic of the continuities of time and space between the two neighbor frames in sequence images,a method based on Kalman filter for predicting position is proposed to match feature points in sequence images rapidly.Four parameters,which are used to describ a similar matrix,are used as state vector in Kalman filter.After four parameters are predicted by Kalman filter,the similarity transformation from the current frame to the next frame can be built,and the correspondence between sub-regions of the current frame and the next frame can be found.Then,all feature points in pairs of corresponding sub-regions are matched respectively.Finally,the state vector including four parameters is updated from an observed similar matrix estimated by all matched corresponding points.The result of an experiment shows that when the magnification of sub-region with the size of 128×128is equal to 1.5,the accuracy of the proposed local matching method is exceed 90%,and the consuming time is decreased by 50%relative to that of global matching method.It illustrates that the proposed matching method can be used effectively to match sequence images.
出处
《光电子.激光》
EI
CAS
CSCD
北大核心
2016年第3期339-346,共8页
Journal of Optoelectronics·Laser
基金
国家"863"计划(2011AA09A102)
浙江理工大学521人才和浙江理工大学科研启动基金(14022211-Y)资助项目
关键词
图像拼接
局部匹配
卡尔曼滤波
相似变换
SURF算子
image stitching
local matching
Kalman filter
similar matrix
speeded up robust features(SURF)descriptor