为解决PBRF-SLAM中由于粒子退化和粒子耗尽而导致的定位失真和建图一致性差的问题,提出了基于海鸥优化和最小方差重采样的优化方法。在PRBF-SLAM的采样过程中,采样一系列辅助粒子,并利用海鸥优化算法对这些粒子进行寻优,找到估计位姿的...为解决PBRF-SLAM中由于粒子退化和粒子耗尽而导致的定位失真和建图一致性差的问题,提出了基于海鸥优化和最小方差重采样的优化方法。在PRBF-SLAM的采样过程中,采样一系列辅助粒子,并利用海鸥优化算法对这些粒子进行寻优,找到估计位姿的最优解,从而避免因陷入局部极值导致的粒子退化。在PRBF-SLAM的重采样过程中,采用最小方差重采样方法替换原先的重采样方法,充分使用辅助粒子,尽可能保证重采样后粒子的多样性。利用Intel Research Lab和ACES Building公开数据集进行SLAM仿真,结果表明优化后的算法相比Gmapping算法总体的平移误差分别降低了36.36%和41.67%,总体的旋转误差分别降低了33.33%和40%。展开更多
RBPF( Rao-Blackwellized Particle Filter) is a popular PF( Particle Filter) in decreasing the dimension of estimation problems and FastSLAM( Fast Simultaneous Localization and Mapping) is a RBPFbased algorithm. In Fast...RBPF( Rao-Blackwellized Particle Filter) is a popular PF( Particle Filter) in decreasing the dimension of estimation problems and FastSLAM( Fast Simultaneous Localization and Mapping) is a RBPFbased algorithm. In FastSLAM,each particle carries a large amount of data which results in low computing efficiency and large memory space occupancy. To solve this problem,a RBPF algorithm with non-intact particle data is studied. The key idea is to differentiate the particle data. Through the screening of particles,the number of particles carrying individual map data is limited to reduce the data occupied space and speed up the computational efficiency. The simulation and experiment results have verified the effectiveness and accuracy of the algorithm. Compared with the original one,this proposed algorithm reduces time consumption by 18%-34% and considerably saves memory space.展开更多
为了解决现有激光SLAM(simultaneous localization and mapping)方法忽略垂直方向漂移而导致的高度不准确和地图重影问题,提出了一种基于垂直约束的紧耦合激光惯性SLAM方法。该方法结合激光雷达传感器的安装高度以及点到激光雷达的距离...为了解决现有激光SLAM(simultaneous localization and mapping)方法忽略垂直方向漂移而导致的高度不准确和地图重影问题,提出了一种基于垂直约束的紧耦合激光惯性SLAM方法。该方法结合激光雷达传感器的安装高度以及点到激光雷达的距离提取精确的地面点,基于提取的地面点设计了一种考虑垂直方向残差的激光里程计,使用两步列文伯格-马夸尔特(Levenberg-Marquardt,L-M)方法来求解姿态变换,这些残差将有助于在垂直方向上收敛到最优解。使用简单有效的基于欧氏距离的回环检测方法避免地图重影问题。为验证算法的优越性,在KITTI数据集及真实场景下均进行了相关实验。在KITTI数据集上,与LeGO-LOAM、LIO-SAM和Point-LIO相比,轨迹均方根误差(root mean square error,RMSE)分别降低了47.62%、33.14%和73.79%。在实测校园环境中,与LeGO-LOAM、LIO-SAM和Point-LIO相比,RMSE分别降低了83.56%、13.55%和82.04%,从而验证了提出方法具有更高的定位精度。展开更多
文摘为解决PBRF-SLAM中由于粒子退化和粒子耗尽而导致的定位失真和建图一致性差的问题,提出了基于海鸥优化和最小方差重采样的优化方法。在PRBF-SLAM的采样过程中,采样一系列辅助粒子,并利用海鸥优化算法对这些粒子进行寻优,找到估计位姿的最优解,从而避免因陷入局部极值导致的粒子退化。在PRBF-SLAM的重采样过程中,采用最小方差重采样方法替换原先的重采样方法,充分使用辅助粒子,尽可能保证重采样后粒子的多样性。利用Intel Research Lab和ACES Building公开数据集进行SLAM仿真,结果表明优化后的算法相比Gmapping算法总体的平移误差分别降低了36.36%和41.67%,总体的旋转误差分别降低了33.33%和40%。
基金Sponsored by the National Natural Science Foundation of China(Grant No.61673125)the Frontier and Key Technology Innovation Special Funds of Guangdong Province(Grant Nos.2016B090910003 and 2015B010917003)+1 种基金the Natural Science Foundation of Guangdong Province(Grant No.2015A030308011)the State International Science and Technology Cooperation Special Items(Grant No.2015DFA11700)
文摘RBPF( Rao-Blackwellized Particle Filter) is a popular PF( Particle Filter) in decreasing the dimension of estimation problems and FastSLAM( Fast Simultaneous Localization and Mapping) is a RBPFbased algorithm. In FastSLAM,each particle carries a large amount of data which results in low computing efficiency and large memory space occupancy. To solve this problem,a RBPF algorithm with non-intact particle data is studied. The key idea is to differentiate the particle data. Through the screening of particles,the number of particles carrying individual map data is limited to reduce the data occupied space and speed up the computational efficiency. The simulation and experiment results have verified the effectiveness and accuracy of the algorithm. Compared with the original one,this proposed algorithm reduces time consumption by 18%-34% and considerably saves memory space.
文摘为了解决现有激光SLAM(simultaneous localization and mapping)方法忽略垂直方向漂移而导致的高度不准确和地图重影问题,提出了一种基于垂直约束的紧耦合激光惯性SLAM方法。该方法结合激光雷达传感器的安装高度以及点到激光雷达的距离提取精确的地面点,基于提取的地面点设计了一种考虑垂直方向残差的激光里程计,使用两步列文伯格-马夸尔特(Levenberg-Marquardt,L-M)方法来求解姿态变换,这些残差将有助于在垂直方向上收敛到最优解。使用简单有效的基于欧氏距离的回环检测方法避免地图重影问题。为验证算法的优越性,在KITTI数据集及真实场景下均进行了相关实验。在KITTI数据集上,与LeGO-LOAM、LIO-SAM和Point-LIO相比,轨迹均方根误差(root mean square error,RMSE)分别降低了47.62%、33.14%和73.79%。在实测校园环境中,与LeGO-LOAM、LIO-SAM和Point-LIO相比,RMSE分别降低了83.56%、13.55%和82.04%,从而验证了提出方法具有更高的定位精度。
