Enhanced Mobile Robot Odometry With Error Kalman Filtering Incorporating 3D Point Cloud Intensity
Xiongwen Jiang,Taiga Kuroiwa,Haolan Zhang,Takato Yoshida,Linfeng Sun,Yu Cao,Haohao Zhang,Takahiro Kawaguchi,Seiji Hashimoto
DOI: https://doi.org/10.1109/access.2024.3434578
IF: 3.9
2024-08-06
IEEE Access
Abstract:Simultaneous localization and mapping (SLAM) based on light detection and ranging (LiDAR) is a trending research direction. This paper proposes a tightly coupled LiDAR inertial odometry (LIO) system that integrates geometric and textural information from the environment, aiming to address the key challenge of autonomous robot navigation. Most existing methods only consider geometric information and assume that reflective surfaces are ideal Lambertian surfaces, neglecting the variable reflective properties of surfaces at different angles and distances. Innovatively, we propose a new LiDAR intensity model that compensates for distortions in intensity values caused by incident angles and distances. Furthermore, we use intensity discontinuities as line features to reflect texture changes in the environment. These points are utilized to calculate residuals of line features, combined with surface features derived from geometric information, and integrated into the observation equations as part of an error-state Kalman filter sensor fusion scheme, thus constructing a tightly coupled LiDAR-inertial odometry system. Compared with the classic LOAM algorithm based on feature extraction and a state-of-the-art algorithm FAST-LIO, the results demonstrate that our system improves trajectory accuracy by approximately 16.1% with LOAM, 15.8% with LOAM+IMU and 11.2% with FAST LIO, rotation accuracy error precision by approximately 15.9% with LOAM, 15.3% with LOAM+IMU and 8.2% with FAST-LIO, demonstrate that proposed method significantly outperforms these methods in terms of localization accuracy and path generation precision.
computer science, information systems,telecommunications,engineering, electrical & electronic