基于双焦单目立体视觉的多层次特征检测算法  

Multilevel Feature Detection Algorithm Based on Bifocal Monocular Stereo Vision

作　　者：刘昕鑫[1] 王元庆[1] 

机构地区：[1]南京大学电子科学与工程系,江苏南京210093

出　　处：《计算机测量与控制》2009年第2期414-417,共4页Computer Measurement &Control

基　　金：国家自然科学基金资助项目(60472026);国家自然科学基金重点项目(60832003)

摘　　要：特征检测是立体视觉技术的重要组成部分;与传统的Canny或Harris等边缘与角点检测相分离的算法不同,采用相位一致性特征检测算法可以同时检测出边缘和角点;相位一致性特征检测在频域进行,与特征点形状及图像的对比度无关,定位效果好,所得角点是边缘点的子集;在特征检测过程中,利用四叉树将图像分割为与图像特征分布相关的多个区域,在各个区域内分别进行处理;同时,通过控制四叉树的分解级数及特征检测的阈值实现了图像的多层次特征检测,实现结果表明,与传统的特征检测算法相比,该算法效果更为理想,且能更好地适用于单目立体视觉系统。Feature detection is important in 3D machine vision, This paper mainly focuses on the feature detection algorithm in stereo vision by monocular system with bifocal lens. Different from the traditional Canny and Harris edge or corner detection, we use the phase congruency feature detection to find the edges and corners. This method operates in frequency domain, is irrelevant to the interest point＇s shape and the contrast of the image, has a good location. Also the corners are a subset of the edge points. We also use the quad tree to divide the image into several parts in each parts we do the feature detection separately. By controlling the threshold and the level of the quad we can get the multilevel feature map. The experimental results show that this method is more ideal than the traditional ones before and is more suitable for the monocular stereo vision.

关 键 词：单目立体视觉 相位一致性 四叉树 多层次 

分 类 号：TN911.73[电子电信—通信与信息系统]