Visual SLAM-Based Autonomous Navigation Of An Unmanned Surface Vehicle

Unmanned surface vehicle with autonomous navigation capabilities and capable of completing tasks in complex water surface environments have become a global research hotspot.Among them,the real-time positioning and mapping(Simultaneous Localization and Mapping,SLAM)technology can provide a real-time environment for autonomous navigation of unmanned surface vehicle.Information and location information are the key to autonomous navigation for unmanned surface vehicle.Based on the research of the visual inertial navigation(VINS-MONO)SLAM algorithm,this thesis proposes a visual inertial navigation(PLVINS-MONO)SLAM algorithm based on the integrated characteristics of points and lines.The algorithm integrates visual information and inertial navigation information to perform real-time navigation environment perception and self-positioning of unmanned surface vehicle,providing key information for autonomous navigation of unmanned surface vehicle.Firstly,in view of the problem of the dynamic area of the water surface in the feature point detection of the unmanned surface vehicle visual image,this paper proposes a front-end design method of the visual odometer based on the HSV(Hue Saturation Value)color area segmentation algorithm.Segmentation of the image obtained by unmanned surface vehicle,excluding the dynamic area of the water surface in the image,to filter out the effective feature points in the world coordinate system in the image,and improve the accuracy and pose estimation of the feature points used by the system in the pose estimation process.Because the line feature information has the invariance of illumination and viewing angle in the water surface environment,its feature information performance is more stable.Therefore,this thesis adds line features to the front end of the visual odometer to enrich the types of feature information extracted by the front end of the visual odometer.Integrate point features and line features,and use point and line feature information to complement each other to improve the stability and robustness of the overall system.Finally,a simulation experiment on the EoRoC data set verifies the effectiveness of the improved visual odometer front end.Through a comparative study with the pure visual SLAM positioning results,the positioning results reflect that the improved visual odometer front end has a more ideal positioning effect,and is closer to the true running trajectory of the data set.Secondly,in order to obtain an accurate global consistent trajectory map,this thesis constructs a visual dictionary based on the comprehensive characteristics of offline points and lines in a water surface environment to improve the accuracy of closed-loop detection.Use 10,000 pictures taken during the actual sailing of the surface ship to train the point-line comprehensive feature visual dictionary.Through in-depth comparison with the point feature visual dictionary and the line feature visual dictionary,it is found that the point-line comprehensive feature visual dictionary proposed in the closed loop detection significantly shortens the query time.The experimental results reflect the superiority of the dot-line comprehensive feature visual dictionary constructed in this thesis.Finally,combining the front-end design of the visual odometer based on the comprehensive features of points and lines and the training of the visual dictionary of the comprehensive features of the points and lines,this thesis proposes a visual inertial navigation SLAM algorithm based on the comprehensive features of points and lines,and systematically designs the visual odometer front-end and system Initialization,back-end optimization and closed-loop detection 4 modules.The visual inertial navigation unmanned surface vehicle data set is used for systematic simulation experiment verification,and the experimental results verify the effectiveness of the improved visual inertial navigation SLAM system under water environment.By making an unmanned surface vehicle data set,evaluating and comparing the pure visual SLAM algorithm,the unimproved VINS-MONO algorithm,and the PIVINS-MONO algorithm proposed in this thesis,the experimental results show that the improved PLVINS-MONO algorithm has a more ideal positioning effect.The positioning trajectory is closer to the real trajectory of unmanned surface vehicle.