Multi-Sensor Fusion SLAM System For Unmanned Vehicles With Loopclosure Constraints

Artificial intelligence is an important technology in the contemporary context.As an important branch of artificial intelligence,computer vision plays an increasingly important role.Simultaneous Localization and Mapping(SLAM)is a branch of computer vision and is mainly used in the robot industry.The main task of the robot is to use the sensors it carries to perceive the surrounding environment in an unknown environment,and to achieve its own positioning and map construction through SLAM technology.At the same time,the development of multi-sensor fusion technology has also made robots use multiple sensors to enhance their performance in positioning and navigation.At present,this technology has been very maturely applied in the driverless car industry,and it has a very wide range of applications in drones,virtual reality,and augmented reality.Aiming at the problems of difficult and inaccurate localization of laser SLAM in large outdoor scenes,this article studies how to improve the localization accuracy of laser SLAM from two aspects.The use of laser point cloud loop detection to reduce the cumulative error of pose estimation and map construction,and multi-sensor fusion technology is used to improve the accuracy and robustness of the laser SLAM system.The main work and results of this article are as follows:This paper first studies the robot SLAM algorithm in large-scale outdoor scenes,and proposes a segment-based loop detection method for the problem of low efficiency and low accuracy of loop detection for point clouds generated by the robot using lidar.The process and implementation of this laser SLAM are introduced,including the extraction of feature segments,pre-processing of point clouds,point cloud segment matching,and pose loop closure.Then,in order to solve the problem that the search space of the point cloud affects the search efficiency,this paper proposes a positionbased candidate segment filtering algorithm.This method can effectively reduce the time required for loop detection and improve the accuracy of loop detection.Finally,this article changed the back-end optimization method to make loop closure fast and effective and reduce map errors.In this paper,experiments are performed on multiple data sets using this method,and the method is proved to be effective.Secondly,this paper studies the improvement of positioning accuracy of the robot,and aims at the problem that the robot uses a single sensor to generate large errors in the positioning and navigation process.This paper proposes a positioning algorithm based on multi-sensor data fusion.First,the wheel odometer and the inertial measurement unit are fused using extended Kalman filtering,and the fused information is used to filter out the wrong GPS information and provide the initial location of the local map construction.NDT algorithm is used in laser localization based on local maps.Finally,we use Kalman filter to fuse GPS and lidar localization result,and use the result as the final pose estimate.Finally,this paper tests the loop detection algorithm based on point cloud segments by building a mobile robot platform.Tests on the Kitti dataset and our datasets,result shows that the SLAM system proposed in this paper has higher accuracy and more robustness in pose estimation and map construction than many SLAM systems.In addition,in this paper,the mobile robot positioning algorithm based on multi-sensor data fusion is tested in the field.The test conditions include the weak GPS information and the complete loss of GPS information.The results show that the SLAM system proposed in this paper can locate well in various extreme cases.