基于改进的DLT算法的PnP问题求解  

Problem-solving of Pn P based on an improved DLT algorithm

作　　者：张培科[1,2] 武元新[2] 

机构地区：[1]中南大学航空航天学院,长沙410083 [2]上海交通大学电子信息与电气工程学院,上海200240

出　　处：《信息技术》2018年第1期36-41,共6页Information Technology

基　　金：国家自然科学基金(61422311);湖南省自然科学杰青基金(2015JJ1021)

摘　　要：摄像机绝对位姿估计问题,即透视n点问题(Perspective-n-point Problem,Pn P)在计算机视觉、摄影测量、机器人技术与增强现实等领域具有重要应用。解决该问题的经典算法直接线性转换(Direct Linear Transform,DLT)算法与其他高精度算法相比计算速度快但估计结果误差较大。文中提出通过充分利用旋转矩阵的约束改进DLT算法,仿真实验测试表明:新算法显著地降低了DLT算法的位姿估计误差,且与高精度算法相当,同时保留了计算效率高的特性。The camera absolute pose problem,i. e. the perspective-n-point problem,has valuable applications in computer vision,photogrammetry,robotics and augmented reality. The basic solution to this problem,the direct linear transform（ DLT）,is fast in computing but has larger estimation error compared with other high accurate methods. An improved DLT algorithm based on the intrinsic constraint of the rotation matrix is proposed in this paper,and the simulation experiments show that the novel approach improves the DLT algorithm accuracy and is comparable in accuracy to those high accurate methods at only a half of computational cost.

关 键 词：透视n点算法 绝对位姿 直接线性转换 计算机视觉 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]