Research On Semantic SLAM Technology Based On Laser Vision Fusion

As a current research hotspot in the field of unmanned driving,combining with the real-time localization and mapping(Simultaneous Localization and Mapping,SLAM)technology of semantic information has a single sensor,limited scenes,low accuracy,and semantic information cannot be combined with the SLAM process.And other issues.This paper proposes a semantic SLAM algorithm based on the fusion of laser and vision.By fusing point cloud information and image data,a semantic map of the environment is constructed.At the same time,the accuracy and robustness of the laser SLAM algorithm is improved based on the semantic information.The main research contributions of this article are:(1)This paper proposes a new type of semantic SLAM framework.By fusing the data collected by lidar and camera,it can obtain richer environmental semantic information under the premise of high-precision depth measurement.Compared with the previous semantic SLAM The algorithm has significant advantages,and the proposed algorithm is verified and tested based on simulation experiments and real vehicle tests.(2)A semantic mapping method integrating visual information is designed,the point cloud is projected to the semantically segmented image to obtain the corresponding semantic information,the semantic map is constructed through point cloud matching,and the corresponding design is designed for the problems of noise and storage.Solution.The semantic map constructed in this paper has the characteristics of rich information,small storage capacity,and high accuracy.Compared with the point cloud map,it is more in line with the law of human cognition and the interaction of the environment,and the subsequent scalability and practicality are stronger.(3)Back-end optimization and loopback detection method based on semantic information is Designed.First,the back-end optimization link is designed by combining semantic information,the semantic segmentation data is associated and corresponding,and geometric information and semantic information are included in the overall optimization process at the same time.Secondly,a loop detection node is designed,which successively searches for possible closed loops through graph matching and vertex matching fused with semantic information,and uses geometric information for fine verification.Using semantic information in this way improves the positioning accuracy of laser SLAM.In addition,for the problem of high real-time requirements of the algorithm,related efficiency optimization methods are designed to improve the overall operating speed;for the characteristics of the semantic point cloud fan-shaped low-view angle,corresponding processing steps are designed to improve the robustness of the algorithm.