基于改进EKF的激光和视觉SLAM融合算法

Laser and Visual SLAM Fusion Algorithm Based on Improved EKF

角点特征在机器人同步定位与建图(Simultaneous Localization and Mapping,SLAM)系统中具有关键性的作用。然而,由于环境差异、机器人运动距离和传感器的影响,导致现有测量方法的角点估计误差较大。本文在原有使用扩展卡尔曼滤波(Extended Kalman Filter,EKF)融合激光和视觉SLAM数据的基础上,引入多新息理论,提出了多新息改进EKF融合激光和视觉SLAM数据算法。由于多新息理论能有效利用历史时刻的数据,使系统在原先只使用当前时刻数据的情况下,扩展为能够利用之前多个时刻的有效数据。因此,利用多新息理论改进EKF,可以充分利用之前时刻由角特征和垂线特征融合成的角点结果,从而提升角点估计精度和建图结果。实验结果表明,在室内坏境中,本文方法在迭代次数20次和100次时平均误差分别为0.0268和0.0109,相较于未改进EKF方法,角点估计的精度平均提升了33.9%。

The corner point feature plays a key role in the simultaneous localization and mapping(SLAM)system of the robot.However,due to the influence of environmental differences,robot moving distance and sensors,the existing methods have the problem of large corner point estimation error.The multi-neo-information theory was introduced,and a multi-neo-information improved EKF fusion laser and visual SLAM data algorithm was proposed based on the original use of extended kalman filter(EKF)to fuse laser and visual SLAM data.Since the multi-neo-information theory could effectively use the data of historical moments,the system,which originally only used the data of the current moment,was expanded to use the data of the previous multiple moments.Therefore,the use of multi-neo-information theory to improve EKF could make full use of the corner point results fused by the corner features and perpendicular features at the previous moment,so as to improve the accuracy of corner point estimation and mapping results.The experimental results show that in the indoor environment,the average errors of the proposed method are 0.0268 and 0.0109 at 20 iterations and 100 iterations,respectively,and the accuracy of corner point estimation is improved by 33.9%compared with the unimproved EKF method.