计算机视觉的PNP问题的最优解  

OPTIMAL SOLUTIONS OF THE PNP PROBLEM IN COMPUTER VISION

作　　者：徐文立[1] 

机构地区：[1]清华大学自动化系,北京100084

出　　处：《自动化学报》1992年第5期522-531,共10页Acta Automatica Sinica

基　　金：国家教委和清华大学资助

摘　　要：本文讨论了计算机视觉的PNP(Perspective-N-Point)问题:根据观察到的n个已知特征点的象点求解被观察物体相对于相机的三维运动参数。由于象噪声,该问题本质上是非线性最优化问题。本文导出一个闭式解,并提出若干克服象噪声影响的方法。仿真试验的结果表明本文的方法有很好的应用前景。This paper gives a comprehensive treatment of the PNP (Perspective-N-Point) problem in computer vision: find the 3-D motion parametersof a viewed object relative to a camera from the observed image points of n known feature points. This problem is essentially one of nonlinear optimization due to the image noise. In this paper, a closed-form solution is developed, and several methods to combat the image noise are presented. The simulation results show that the approaches given in this paper hold great promise in application.

关 键 词：最优解 机器视觉 PNP问题 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]