一种基于多模式单演特征检测与匹配的三维视觉测量方法  

Three-Dimensional Vision Measuring Technique Based on Multi-Modal Monogenic Features Detecting and Matching

作　　者：李进军[1,2] 赵宏[2] 

机构地区：[1]第二炮兵工程学院,陕西西安710025 [2]西安交通大学机械制造系统工程国家重点实验室,陕西西安710049

出　　处：《光学学报》2011年第7期140-148,共9页Acta Optica Sinica

基　　金：国家863计划(2008AA04Z121);国家自然科学基金(50575169);陕西省西安市科技创新支撑计划(2010CXY1007)资助课题

摘　　要：三维视觉测量技术从一幅或多幅二维图像或者视频剪辑自动获取现实世界中目标和场景的几何模型、运动和形变等信息,从而识别和理解三维现实环境。针对传统三维视觉测量易受亮度变化、尺度变化、旋转变化及噪声影响的缺点,提出了一种基于多尺度单演特征检测与匹配的三维视觉测量新方法。该方法采用泊松(Poisson)滤波算子和里斯(Riesz)变换提取图像的多模式单演特征,包括局部幅值、局部方向、局部相位和彩色相位;模拟人眼视觉功能构建多模态的局部特征描述子及特征空间的相似性距离测量函数;将该相似性测量应用于立体匹配和三维视觉重建中,并进行了定性和定量的实验评估。Three-dimensional vision measuring technique aims at automatically reconstructing the properties of object and scene such as shape,motion,and deformation from one or more images or video clips in the real world around us,so that three-dimensional true environment can be recognized and understood.To solve the problems such as brightness change,scale change,rotation change and noise in conventional approaches,a new three-dimensional vision measuring technique is proposed.Firstly,the multi-modal isotropic monogenic features which include geometric feature（local orientation）,structure feature（local amplitude and local phase）,color phase and color texture are extracted by Poisson kernel and Riesz transforms.Secondly,a novel multi-modal feature descriptor and a similarity measurement function are constructed by simulating human vision function.Finally,the proposed similarity measurement function is applied to stereo matching and three-dimensional vision reconstruction.At the same time,the qualitative and quantitative evaluations are finished on the synthetic and natural stereo images.

关 键 词：三维视觉测量 单演信号分析 立体匹配 

分 类 号：TH741[机械工程—光学工程]