惯导/双目视觉位姿估计算法研究  

Research on SINS/binocular vision integrated position and attitude estimation algorithm

作　　者：于永军[1] 徐锦法[1] 张梁[1] 熊智[2] 

机构地区：[1]南京航空航天大学直升机旋翼动力国家级重点实验室,南京210016 [2]南京航空航天大学导航研究中心,南京210016

出　　处：《仪器仪表学报》2014年第10期2170-2176,共7页Chinese Journal of Scientific Instrument

基　　金：中国博士后科学基金(2013M541668);国家自然科学基金(61374115);江苏高校优势学科建设工程资助项目

摘　　要：无人飞行器自主导航系统作为无人机自主控制的核心,是制约无人机性能提升的关键因素。传统的惯导/GPS组合导航系统受其自主性、稳定性等限制无法满足无人机导航需求。在分析视觉导航特性的基础上,针对无人飞行器复杂运动特性,设计了基于非线性尺度空间的双目视觉三维位姿估计算法。在此基础上,针对双目视觉导航存在的位姿估计误差累积问题,提出了基于惯导/视觉融合的无人飞行器位置和姿态估计算法,利用卡尔曼滤波实现惯导与视觉信息的融合。通过仿真验证和实际跑车验证结果表明,提出的双目视觉位姿估计算法相比SIFT算法能够有效提高视觉图像特征检测准确度,惯导/视觉组合算法能够有效提高位姿估计精度,组合定位精度优于2.5 m。Autonomous navigation system as the core of unmanned aerial vehicle （ UAV ） autonomous control is the key factor restricting UAV performance upgrade. Due to the constraint of autonomy and stability, traditional SINS/GPS integrated navigation system can not meet the navigation requirement of UAV. Based on the analysis of vision navigation characteristics, aiming at the complex dynamic char- acteristics of UAV, the binocular vision 3D position and attitude estimation algorithm based on nonlinear scale space is designed. To solve the problem of error accumulation of binocular vision navigation, a SINS/vision integrated position and attitude estimation algorithm is proposed; Kalman filter is used to fuse the SINS and vision information. Simulation verification and car-based test results indicate that compared with SIFT algorithm, the proposed binocular vision algorithm can effectively improve the detection accuracy of visual images, the SINS/vision integrated algorithm can effectively improve the position and attitude estimation accuracy; and the position accuracy is better than 2.5 m.

关 键 词：捷联惯导导航 双目视觉导航 非线性尺度空间 组合导航 卡尔曼滤波 

分 类 号：TH761.7[机械工程—仪器科学与技术]