基于改进SURF算法的移动目标实时图像配准方法研究Studyon.PDF

第 38 卷第 1 期 通 信 学 报 Vol.38 No.1 2017 年 1 月 Journal on Communications January 2017 doi:10.11959/j.issn.1000-436x.2017020 基于改进 SURF 算法的移动目标实时图像配准方法研究 巨刚，袁亮，刘小月，岳昊恩 （新疆大学机械工程学院，新疆乌鲁木齐 830047 ） 摘 要：针对目标在移动过程中实时视觉图像特征点提取与配准的不稳定性，提出一种多算法融合的改进配准方 法。首先，采用双边滤波、Canny 边缘检测及形态学处理方法得到具有较强顽健性特征的边缘周边检测区域并基 于离散 Gaussian-Hermite 矩对 SURF 算法中的 Hessian 矩阵进行修正，重新定义特征描述向量，同时采用肯德尔 系数对配准的特征点进行约束。其次，通过融合光谱辐射颜色不变量模型及 I_SURF 算法对实时视觉彩色图像进 行配准。最后，将改进算法与目标自适应更新算法相结合，实现了移动目标在室内环境中的实时匹配。实验结果 表明，在不同旋转尺度下，改进算法的静态图像配准较 SURF 算法具有较高配准精度，移动图像特征点提取及配 准数量的稳定性达到 97% 以上。 关键词：改进 SURF；移动目标；图像配准；实时性；Gaussian-Hermite 矩 中图分类号：TN911.73 文献标识码：A Study on mobile target real-time image registration based on improved SURF algorithm JU Gang, YUAN Liang, LIU Xiao-yue, YUE Hao-en (School of Mechanical Engineering, Xinjiang University, Urumqi 830047, China) Abstract: For the unstability of the real-time visual image characteristic point extraction and matching for the mobile target, an improved registration method of the multiple algorithm-fusion was introduced. Firstly, the method of bilateral filtering, Canny edge detection and morphological processing, was adopted to get the more robust images edge map. And the founded points were limited in this edge map, then the Hessian matrix of SURF based on the discrete Gaus- sian-Hermite moment was modified. The character description vector was redefined in the algorithm. Following the above analysis, the Kendall coefficient co