景象匹配导航中的航路点自动提取算法  

Automatic Extraction Algorithm of Landmarks for Scene-Matching-Based Navigation

作　　者：公续平 程咏梅[1] 宋琳[1] 刘楠[1] 

机构地区：[1]西北工业大学自动化学院,陕西西安710072

出　　处：《计算机仿真》2014年第8期60-63,203,共5页Computer Simulation

基　　金：西安市科技计划项目(CXY1350(2))

摘　　要：针对景象匹配导航中适配区提取算法选取适配区时存在匹配精度低、速度慢的问题,提出了一种航路点自动提取算法。利用图像灰度信息、边缘信息和图像间互相关信息,对航拍图像中不同尺度图像样本进行参数训练,进而得到各尺度下的分类参数,并通过支持向量机完成航路点区域的自动粗分类。针对航路点选取过于密集导致选取的航路点不唯一及计算量增加的问题,在欧式距离分类法基础上,提出了多分类无监督聚类算法,从而实现航路点的优化。仿真和航拍数据结果表明改进的算法可以自动提取航路点,并且提取的航路点具有信息量丰富、稳定性好、匹配精度高及速度快等优点。A method for automatic extraction of landmarks for vision - based navigation is proposed in this paper to solve the short- comings of low matching probability and slow speed in scene matching area＇s selection methods. According to different size of sample images, this algorithm uses image gray features, edge features and correlation plane statistic features to training parameters via Support Vector Machine （SVM）. Then we use SVM to classify tes- ting images, and finally we take unsupervised clustering algorithm to optimize landmarks selected by SVM. Simulation experiments and real flight data show that our method can automatic extract landmarks, and the extracted land- marks have characters of stability, high matching accuracy and short matching time.

关 键 词：航路点 支持向量机 图像特征 无监督分类算法 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]