Research And Application Of Unmanned Vehicle Map Construction Method Based On Multi-Source Information Fusion

In recent years,autonomous driving technology has gradually become a focus of attention in the automotive industry and academic community.As a key component of autonomous driving technology,autonomous driving maps can provide good data support for functions such as positioning and environment perception of intelligent unmanned vehicles.Among them,the integration of semantic information in autonomous driving maps can further improve the interaction between unmanned vehicles and the environment.The updating ability of autonomous driving maps for dynamic changes in the environment also has a great impact on the safety of unmanned vehicle driving.Based on the above analysis,this thesis combines Simultaneous Localization and Mapping(SLAM)algorithm with semantic segmentation network to construct a semantic point cloud map,and uses Octree and improved ray casting method to effectively update the point cloud map.The main research contents of this thesis are as follows:(1)Research on the hardware system,software system composition,and the overall framework of the semantic map construction system of the mobile experimental platform.Research on the spatial reference definition and transformation relationship of the system.Research on the offline calibration scheme of the laser radar and camera based on the rectangular calibration board,and complete the spatial calibration between the Livox Avia laser radar and the camera based on the scanning characteristics of the laser radar.Detailed derivation of the principle of inertial measurement unit(IMU)inertial solution.(2)Research on SLAM algorithm based on data fusion of laser radar and IMU.Study the classical Bayesian filtering idea,Kalman filtering theory,and error-statebased filtering method in state estimation,and derive the iterative error-state Kalman filtering(IESKF)method.Establish the error state propagation equation of the IMU,use the principal component analysis method to fit the plane for element matching to establish the observation equation,and use IESKF to fuse laser radar and IMU data to realize pose estimation and point cloud map construction.Experimental results verify that the fusion positioning scheme of laser-inertial data can improve the positioning accuracy during high-speed motion and ensure the stability of the system.(3)Research on point cloud map construction method containing color and semantic information.Research on the co-visibility extraction method of laser radar and camera to reduce redundant data and realize the construction of a color point cloud map.Based on the DeepLabV3+ semantic segmentation network,the eight semantic information of pole-like objects,roads,cars,vegetation,buildings,bicycles,pedestrians,and signs in the campus scene are segmented.Propose a weighted optimization algorithm based on point cloud geometric features and image semantic features to optimize semantic information and achieve accurate semantic map construction.(4)Research on the updating method of the point cloud map.Research on the Octree data structure,take voxel as the minimum update unit,and research on the voxel grid confidence description method based on static binary Bayesian filtering.Propose an improved ray casting method to construct virtual points to solve the problem of the failure of the traditional ray casting method caused by the extraction of the co-visibility area and improve the updating accuracy of the point cloud map.The semantic map construction system proposed in this thesis can construct a point cloud map with rich semantic information while ensuring its own pose accuracy.At the same time,the system has the ability of dynamic updating,which can avoid the problem of map timeliness and improve the ability of intelligent driving vehicles to understand the surrounding environment.The system can provide reliable data support for functions such as positioning,environment perception,and decision-making of intelligent driving vehicles,thereby improving the efficiency and reliability of autonomous driving functions in actual operation.