一种基于历史特征的SURF改进算法研究.docVIP

一种基于历史特征的SURF改进算法研究 　　摘 要： 图像配准是数字图像处理深度应用的基础之一，其中基于SURF的图像配准算法因识别率高而得到广泛的研究与应用，但其数据量大且对计算要求较高，因此提出一种基于对象关联的配准方法，在SURF前端提取对象ROI以检测是否有新的对象进入检测区域从而将新旧对象分为两类分别处理，对于已经存在的对象可根据运动特征关联进行进一步过滤，较大幅度地减少重复特征点的检测和计算，也可避免依赖局部区域像素的梯度方向造成过大的误差。实验结果表明，改进的算法提高了配准率，减少了约20%的计算量，帧率下降至0.8左右时趋于稳定，保证了较好的实时性。 　　关键词： SURF算法； Hessian矩阵； 运动对象识别； 匹配率 　　中图分类号： TN911.73?34 文献标识码： A 文章编号： 1004?373X（2017）03?0038?05 　　Research on a SURF improved algorithm based on historical characteristics 　　HUANG Jinyong1， LI Zhe1， 2， ZHANG Tianfan1， 2 　　（1. College of Technology， Hubei Engineering University， Xiaogan 432000， China； 　　2. School of Automation， Northwestern Polytechnical University， Xi’an 710072， China） 　　Abstract： The image registration is one of the fundaments for further application of the digital image processing. The SURF?based image registration algorithm is widely researched and applied due to its high recognition rate， but it has large data quantity and high calculation requirement， so a registration method based on object association is proposed. The object ROI is extracted in the front end of SURF to detect whether a new object has entered into the detection area， so as to divide the old and new objects into two classes for processing. The existing objects can be further filtered according to the motion feature association to significantly reduce the detection and calculation of the repeated feature points， and avoid the excessive error caused by relying on the gradient direction of the local region pixel. The experimental results show that the improved algorithm has improved the registration rate， its calculation quantity is reduced by 20%， its frame rate trends to be stable when it is dropped to about 0.8， and the algorithm insures good real?time performance. 　　Keywords： SURF algorithm； Hessian matrix； moving object recognition； matching rate 　　0 引 言 　　运动目标识别一直是机器视觉领域的研究热点。它已被广泛应用于视频监控、测绘、机器人导航等领域中。SURF（Speeded Up Robust Features）[1] 是近年应用广泛的一种匹配算法，采用了Harr特征以及积分图像概念，使得同等条件下标准