In the context of autonomous driving systems, SLAM and dynamic object tracking represent pivotal challenges. Autonomous driving scenarios frequently demand the simultaneous acquisition of ego-pose and comprehensive motion information from the surrounding environment to enhance decision-making and scene comprehension.Given the inherent interdependence between these two challenges, a viable approach is to integrate SLAM and object tracking into an interconnected system referred to as SLAMMOT. However, many conventional SLAMMOT solutions rely on a single motion model for object tracking, which may inadequately capture complicated dynamics of real-world motions. In practice, object motion patterns can change from time to time, not conforming neatly to a single model. To handle existing challenges, this paper proposes the IMM-SLAMMOT, a tightly-coupled LiDAR-based SLAMMOT system that utilizes instance semantic segmentation and IMM modelling for dynamic object tracking. Ego-pose and dynamic object states are jointly optimized in an innovative graph optimization framework intimately integrated with the IMM algorithm. Comparative analysis against our baseline, which employs a single motion model for object tracking, demonstrates that the IMM-SLAMMOT outperforms at motion-pattern-transition moments and consistently achieves competitive results in SLAM and multi-object tracking tasks throughout the entire trajectory.
IMM-SLAMMOT: Tightly-Coupled SLAM and IMM-Based Multi-Object Tracking
IEEE Transactions on Intelligent Vehicles ; 9 , 2 ; 3964-3974
2024-02-01
3610742 byte
Aufsatz (Zeitschrift)
Elektronische Ressource
Englisch
Efficient and Accurate Tightly-Coupled Visual-Lidar SLAM
IEEE | 2022
|A tightly coupled LIDAR-IMU SLAM in off-road environment
IEEE | 2019
|Ultra-Tightly Coupled GPS/INS Tracking Performance
AIAA | 2003
|