单目视觉下基于对偶四元数的运动目标位姿确定  

Pose Estimation of Moving Object based-on Dual Quaternion from Monocular Camera

作　　者：冯国虎[1] 章大勇[1] 吴文启[1] 

机构地区：[1]国防科技大学机电工程与自动化学院,长沙市砚瓦池街47号410073

出　　处：《武汉大学学报（信息科学版）》2010年第10期1147-1150,共4页Geomatics and Information Science of Wuhan University

基　　金：国家教育部高等学校博士学科点专项科研基金资助项目(20069998009)

摘　　要：利用对偶四元数研究运动载体和目标之间相对运动的单目算法大多是仿真验证,缺乏实验验证。针对这些不足,在James方法的基础上,扩展了状态量,推导了相应的系统模型和观测模型,组成新的滤波方程。匀速旋转的转台实验验证了改进算法的优越性。In order to estimate position and orientation of object,an EKF method based on dual quaternion from monocular camera was investigated.An improved EKF method based on James method was proposed.In the EKF,relative translation and rotation second derivations of the object were added as state assignment.Object lines are measured variables,which were transformed between coordinate frames by dual quaternion.Corresponding system model and measurement model were deduced.At the same time,some mistakes in James method were rectified.Experiment tests using turntable to demonstrate relative motion and position were done.Actual experimental data of rotation at a constant speed of 20 /s show that the improved method get almost the same precision of pose estimation as James method.While actual experimental data of 40 /s show that the improved method can get better precision and robustness.

关 键 词：单目视觉 对偶四元数 扩展卡尔曼滤波 位姿确定 

分 类 号：P228.41[天文地球—大地测量学与测量工程]