计算机视觉中投影不变量稳定性的分析  

Stability analysis on projective invariants in computer vision

作　　者：孙九爱[1] 吕东辉[2] 庄天戈[1] 

机构地区：[1]上海交通大学生物医学工程系,上海200030 [2]上海大学通信与信息工程学院,上海200072

出　　处：《红外与激光工程》2001年第3期190-192,共3页Infrared and Laser Engineering

基　　金：中法科技合作项目A类第 87号 ;上海市科学发展基金资助项目 (9851 0 70 1 6)

摘　　要：投影不变量具有与视点和摄像机内外参数无关等优点 ,正逐渐成为计算机视觉研究的常用工具。在构造模型库时 ,投影不变量的稳定性很重要。它与图像的质量、特征提取算子的精度和数值计算的精度等有关。首先对共线 4点和共面 5点构成的交比不变量的稳定性进行了分析 ,得出了影响稳定性的因素 ,然后提出在构造目标识别系统的模型时应注意的约束条件 ,实验表明 :在这些条件下构造模型的抗噪声能力大大提高。Projective invariants are independent of viewpoints, intrinsic and extrinsic camera parameters and common used in computer vision gradually. The stability of the projective invariants is very important for the construction of the model library. It is related to the quality of the images, the precision of the feature extractor and the numerical computation. In this paper, cross ratios of the four collinear points and five coplanar points were analyzed first, then some affecting factors were concluded, and some constraints were proposed on constructing model-based object recognition system. The experiments show that the stability of the projective invariants designed under these constrains is improved highly.

关 键 词：投影不变量 交比 稳定性 目标识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]