基于不确定性分析的视觉里程计优化

Optimization of visual odometry based on uncertainty analysis

通过推导视觉里程计中运动参数估计的不确定度,分析了视觉里程计的定位精度.采用矩阵扰动理论,准确计算了基于最小二乘法运动估计算法给出的6个自由度运动参数估计的不确定性,此方法的计算复杂度为O(1).采用扩展卡尔曼滤波器对视觉里程计和惯性测量单元数据进行融合优化,获得了更加准确的机器人定位和姿态信息.融合实验结果表明,融合后的闭合误差比单一的视觉里程计闭合误差减少近49.5%.

The accuracy of visual odometry （VO） was analyzed by deriving the uncertainty of the VO＇s mo- tion estimation. The proposed uncertainty analysis method, based on matrix perturbation theory, esti- mates the uncertainty of motion parameters in six degrees of freedom （DOF） those are computed by least- squares-method. The computational complexity is O（1） . Then an extended Kalman filter （EKF） is used for fusing VO and inertial measurement unit （IMU） data based on above analysis results to obtain more ac- curate position and attitude values of the robot. Experimental results show that the closed-loop error is re- duced by 49.5 % at most when the uncertainty data fusion with IMU is applied to VO.