基于自适应衰减记忆法的视觉惯性里程计算法

Improved Visual Inertial Odometry with Fading Memory Filter

针对机器人快速运动导致系统先验噪声统计特性与实际不符,严重影响视觉惯性里程计定位精度的问题,提出了一种基于自适应衰减记忆法的视觉惯性里程计算法。利用机器人运动状态方程模糊推理得到衰减因子,实现一种新的选取衰减因子的方法;然后利用自适应衰减因子实时修正噪声模型,抑制滤波发散;最后在公开数据集和真实环境下进行实验,该方法整体定位误差在0.6 m内。实验结果表明,所提算法在机器人快速运动时能有效避免滤波发散,对机器人定位性能提升显著。

In order to solve the problem that the statistical characteristics of the prior noise of the system are inconsistent with the reality due to the rapid movement of the robot,which seriously affects the positioning accuracy of the visual inertia odometry,a visual inertia odometry algorithm based on adaptive fading memory method was proposed.The fading factor is obtained by fuzzy inference of robot motion equation of state,and a new method of selecting fading factor is realized.Then the adaptive fading factor is used to modify the noise model in real time to suppress the filter divergence.Finally,experiments are carried out in the open dataset and real environment,and the overall positioning error of this method is within 0.6 m.Experimental results show that the proposed algorithm can effectively avoid filtering divergence when the robot is moving fast,and significantly improve the positioning performance of the robot.