Research On Multi Sensor Fusion Simultaneous Localization Mapping And Multi Object Tracking

Simultaneous Localization and Mapping(SLAM)is an important research problem in the field of robotics.Its basic requirement is for a robot to estimate its own pose and construct an environmental map using various sensor data in an unknown environment.However,classical SLAM theories are based on the assumption of a static environment.When a robot operates in a complex and dynamic real-world environment,the presence of moving objects in the scene can affect the accuracy of localization and the quality of the map,resulting in trajectory drift and ghosting phenomena,which pose challenges to existing SLAM solutions.The purpose of this study is to address the impact of dynamic objects on SLAM in real-world environments.Multiple methods are proposed to enhance the localization capability of SLAM in dynamic environments,reduce trajectory errors,and improve the quality of the constructed map.Specifically,this thesis presents a solution called Multi Sensor Fusion Simultaneous Localization Mapping and Multi Object Tracking(SLAMMOT),which aims to handle the challenges posed by dynamic environments.It tackles the complex problem of simultaneously estimating robot pose,environment map,and dynamic object poses in dynamic environments.The main contributions of this paper are as follows(1)Proposed a visual-lidar fusion SLAM algorithm based on depth completion..First,project the lidar onto the camera plane,use depth completion technology to complete sparse depth,use dense depth and image to implement visual odometry,use lidar point clouds to implement lidar odometry,and use factor graphs to fuse visual-lidar odometry to output self-pose and construct the environment map.(2)Proposed an iterative dynamic registration algorithm that combines self-motion estimation and 3D motion object detection.Based on 3D object detection to obtain the position and size of the object,estimate frame-to-frame motion using point cloud registration,segment and estimate moving objects using estimated poses,remove moving objects from the point cloud,and retain stationary objects to achieve frame-stable point cloud registration in dynamic environments using static environment point clouds.(3)Proposed a tightly coupled simultaneous localization and mapping(SLAM)algorithm with multi-object tracking.Remove the ground from the input point cloud to maintain stable estimation,then combine with the dynamic registration algorithm and design the lidar odometry algorithm.Use the joint probabilistic data association filter to achieve SLAMMOT in a loosely coupled manner,and use global nearest neighbor algorithm to achieve tightly coupled SLAMMOT with factor graphs.Finally,output the environment map,ego-trajectory,and moving and static object trajectories.In order to evaluate the performance of the proposed approaches in this paper,their effectiveness will be validated by comparing their performance with traditional SLAM methods in dynamic environments using publicly available datasets.The following metrics will be used for evaluation: Absolute Trajectory Error(ATE),Relative Pose Error(RPE),and point cloud map Accuracy.Experimental results on multiple sequences demonstrate that the proposed method can achieve accurate localization,and generate precise maps and robot trajectories in dynamic environments.The proposed method outperforms the existing approaches regarding localization accuracy and map quality in various environments,enhancing the robot’s autonomous perception and navigation capabilities.