Core Concepts
Proposing a novel semantic SLAM approach, SD-SLAM, for dynamic scenes using LiDAR point clouds to enhance localization and mapping performance.
Abstract
Introduction to SD-SLAM for dynamic scenes based on LiDAR point clouds.
Addressing challenges of dynamic objects in point cloud maps.
Three main contributions of the proposed SD-SLAM approach.
Evaluation using KITTI odometry dataset showing improved localization and mapping performance.
Detailed methodology including point cloud instance segmentation, preliminary pose estimation, landmark motion state identification, precise pose estimation, loop closure, and mapping.
Results of tests with KITTI datasets showcasing superior performance in vehicle localization and dynamic landmark detection.
Stats
LiDAR point clouds are commonly used for localization and navigation by autonomous vehicles and robots.
Results demonstrate that SD-SLAM effectively mitigates adverse effects of dynamic objects on SLAM.
SD-SLAM achieved the highest localization accuracy across most test sequences.
SD-SLAM provided the best localization performance with loop closure detection on-board.
Quotes
"Results demonstrate that the proposed SD-SLAM effectively mitigates the adverse effects of dynamic objects on SLAM."
"SD-SLAM achieved the highest localization accuracy across most test sequences."