| In recent years,the government encouraged the combined development of industry,medical,education,entertainment with VR/AR,robotics.So,(Simultaneous Localization and Mapping)(SLAM),as the key technology,is facing great challenges.Especially,the problems of real-time,map building accuracy in dynamic environment and limited usage scenarios due to single sensor became the research focus in the field.To address these problems,this paper studies and implements a multi-sensor SLAM system based on vision,inertial measurement units and Li DAR,which not only enables efficient map building and localization in dynamic environments,but also provides reliable positional information in vision failure scenarios.The main research contents of the paper include.(1)A multi-sensor hardware platform which can be adopted by indoor and outdoor environments is built based on depth cameras,inertial measurement units,Li DAR and other sensors to provide an experimental basis for theoretical research.Moreover,since the platform has good platform portability characteristics,it can be installed to a variety of robotic platforms.(2)To address the real-time problem of SLAM system performance caused by the feature extraction and word conversion,this paper propose a parallel scheme of feature processing and pose calculation,which achieved the creation of frame objects by using feature processing threads for feature point extraction and word conversion of bag-of-words model.The experimental results show that the real-time performance of SLAM system can be improved almost 30% by the proposed scheme.(3)To improve the utilization ratio of inter-frame matching information,a calculation framework based on the frame and local map matching information is proposed to reduces the probability of tracking failure due to unqualified inter-frame matching and improves the robustness of the system by combining with inertial measurement units.Experiment results show that the trajectory accuracy root-mean-square error(RMSE)of the proposed framework is 2 cm.(4)To improve the map building accurate of SLAM system in the scene with dynamic objects,the YOLOV5 is adopted to eliminate the feature points and point clouds in the dynamic object detection frame.The experimental results show that the proposed scheme can map the environment with dynamic objects effectively.(5)Considering the demand of SLAM application scenarios in the region where vision methods fail such as no illumination and weak texture,a laser odometer plan is designed based on the idea of real-time Livox laser odometer and map building system to maintain the system pose by combining the visual inertial odometer.This plan not only extends the application scenarios but also improves the practical value of SLAM technology. |
