基于改进克隆卡尔曼滤波的视觉惯性里程计实现  

Implement of visual-inertial odometry based on improved stochastic cloning kalman filtering

作　　者：徐之航 欧阳威 武元新[1] XU Zhi-hang;OUYANG Wei;WU Yuan-xin(Shanghai Key Laboratory of Navigation and Location-based Services,Shanghai Jiaotong University,Shanghai 200240,China)

机构地区：[1]上海交通大学上海市北斗导航与位置服务重点实验室,上海200240

出　　处：《信息技术》2020年第5期21-27,36,共8页Information Technology

基　　金：国家自然科学基金(61673263)。

摘　　要：视觉-惯性多传感器融合方案可以有效解决GNSS据止环境下的导航问题。为了更有效地处理视觉的相对观测信息,文中研究并实现了一种基于克隆卡尔曼滤波(stochastic cloning Kalman Filter,SCKF)与滑动窗方法相结合的低成本视觉惯性里程计紧耦合算法。开源数据集实验表明,文中算法精度超过了多状态约束卡尔曼滤波(MSCKF)开源代码的效果。Kalman filter based visual-inertial multi-sensor fusion scheme can effectively solve the navigation problem in GNSS-denied environment.In order to deal with the relative measurement,a low-cost,sliding-window tight-coupled algorithm of visual-inertial odometry is researched and implemented based on the stochastic cloning Kalman Filter(SCKF).Experiments with open source datasets show that the proposed algorithm has better accuracy than the multi-state constraint Kalman filter(MSCKF) open source code.

关 键 词：惯性视觉里程计 克隆卡尔曼滤波 滑动窗 多状态约束 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]