基于视觉传感器的动态目标位姿估计  

Moving target pose estimation based on visual sensor

作　　者：庞晨涛 顾寄南[1] Pang Chentao;Gu Jinan(Research Center of Mechanical Information,Jiangsu University,Zhenjiang 212013,China)

机构地区：[1]江苏大学制造业信息化研究中心,镇江212013

出　　处：《电子测量技术》2020年第18期107-111,共5页Electronic Measurement Technology

基　　金：江苏省科技成果转化专项资金项目(BA2015026)资助。

摘　　要：针对动态目标位姿估计时容易受到系统噪声影响的问题,提出了摄影测量与自适应无迹卡尔曼滤波算法相结合的位姿估计方法。首先根据目标的运动学模型建立系统的状态空间模型,然后使用摄影测量法求解目标当前时刻的位姿,并结合自适应无迹卡尔曼滤波对目标位姿进行后验估计。考虑到动态目标位姿估计系统中噪声统计特征未知的情况,采用基于极大后验估计的噪声估计器,对系统噪声统计特征进行在线估计。最后通过实验验证了该方法的实用性,对比直接使用摄影测量或者无迹卡尔曼滤波,位姿估计精度显著提高,并实现了系统噪声统计特征的在线估计。A pose estimation method combining photogrammetry and adaptive unscented Kalman filter(AUKF) is proposed to solve the problem that the pose estimation of moving targets are easily affected by noise. Firstly, the system’s state space model was established according to the target’s kinematics model. Then, the photogrammetry method was used to solve the target’s pose at the current moment, and combining with AUKF to estimate the target pose. Considering the situation that the noise statistic characteristics are unknown in the moving target pose estimation system, a noise estimator based on maximum a posterior(MAP) was used to estimate the statistical characteristics of the system noise online. Finally, the practicability of the method is verified through experiments. Compared with using photogrammetry or unscented Kalman filtering directly, the accuracy of pose estimation is significantly improved, and the online estimation of statistical characteristics of system noise is realized.

关 键 词：单目视觉 摄影测量 无迹卡尔曼滤波 噪声估计器 位姿估计 

分 类 号：TP212.9[自动化与计算机技术—检测技术与自动化装置]