Research On SLAM Algorithm Based On Multi-Sensor Fusion

Autonomous driving becomes increasingly popular in the field of artificial intelligence,driven by advancements in related technologies.Autonomous driving gradually draws people’s attention due to its broad market prospect and huge development potential,thereby moving itself from the backstage to the front stage.Simultaneous Localization and Mapping(SLAM),as an important technology in the field of autonomous driving,attracts the attention of experts and scholars at home and abroad because of its ability to perform mapping and localization in unknown environments.When a SLAM system is built using a single sensor,the accuracy of estimating the pose(position and orientation)of the moving vehicle varies due to the limited amount of data,which in turn affects the accuracy and quality of the map.At the same time,the observation error of a single sensor will cause the system to have a wrong estimation of the motion carrier’s pose.Thus,the robustness of the system cannot be guaranteed.However,the multi-sensor fusion SLAM system uses different sensors to collect data and estimate the vehicle’s own motion state,which can give full play to the data characteristics of different sensors achieving effective and accurate estimation of the carrier’s pose.A multi-sensor fusion-based SLAM system can offer advantages over a single-sensor system by mitigating observation errors,thereby enhancing overall system robustness.This is due to the system’s ability to incorporate data from multiple sensors and compensate for discrepancies.Therefore,the SLAM algorithm based on multi-sensor fusion can effectively improve the mapping accuracy,which further provides a safe and effective guarantee for autonomous driving.This paper focuses on the study of multi-sensor fusion-based SLAM map building algorithms,and makes the following contributions on the framework of traditional SLAM algorithms.(1)A multi-sensor fusion SLAM mapping framework based on factor graphs is proposed,aiming to address the two issues associated with single-sensor-based solutions.Specifically,the proposed framework employs front-end odometry factors,IMU preintegration factors,and loop closure detection factors based on the correlation domain to add pose constraints.(2)Regarding the widely existing problem of accumulated errors in long-range mapping using SLAM algorithms,this paper proposes a novel loop closure detection method.A new point cloud global descriptor is designed,which optimizes the matching dimension of the descriptor,resulting in the shortening of the matching time.It further enhances the accuracy and validity of loop closure detection.(3)To address the manual or semi-automatic process required for traditional largescale map stitching,this paper proposes a multi-bag merging mapping mechanism based on loop closure detection.The mechanism works during the factor graph-based back-end optimization process,enabling automatic map stitching.(4)This paper utilized a laboratory data acquisition platform to construct a campus dataset for providing an offline experimental environment.Through conducting loop closure detection experiments,the performance of SLAM algorithm experiments and multi-bag merge mapping experiments on both KITTI and the campus dataset is verified.