单目视觉定位中SURF算法参数的优化  

Optimization of SURF Parameters in Monocular Visual Odometry

作　　者：赵立双[1] 冯莹[1] 曹毓[1] 

机构地区：[1]国防科学技术大学光电科学与工程学院,湖南长沙410073

出　　处：《计算机技术与发展》2012年第6期6-9,共4页Computer Technology and Development

基　　金：中国人民解放军总后勤部资助项目(BY2008J018)

摘　　要：为了提升单目视觉定位方法的定位效率,在基于SURF算法的单目视觉定位系统上对SURF算法参数的选取进行了优化。首先分析了路面图像的特点及路面图像中SURF特征点的特性,据此选取了SURF算法中组数和层数这两个重要参数;其次分析了路面序列图像中特征点数目与hessian矩阵行列式阈值之间的关系,提出了hessian矩阵行列式阈值动态设定方法。通过对SURF算法参数的优化,有效降低了程序的运算量。实验结果表明,该方法能较好满足路面环境下定位的要求,在保证算法精度和稳定性的同时,大幅提高了程序的效率。In order to increase the localization frequency of monocular visual odometry, optimization of SURF parameters was done on monocular visual odometry system based on SURF. Firstly, features of pavement image and the SURF keypoint in it were analyzed, and octave number and octave layer number of SURF were properly chosen. Secondly, the relation between SURF keypoint number and hessian threshold in image sequence was analyzed,and a method of hessian threshold dynamic setting was put forward. By optimizing SURF ＂parameter,the amount of computation was reduced efficiently. The experimental results demonstrate that the method can fit the demand of localization on the pavement, meanwhile, the efficiency of monocular visual odometry is greatly improved as the accuracy and stability are well kept.

关 键 词：机器视觉 单目视觉定位 SURF参数 hessian矩阵行列式阈值 

分 类 号：TP212.14[自动化与计算机技术—检测技术与自动化装置]