Research On Autonomous UAV Localization Based On Visual Semantic Map Method

At present,the application of UAVs in various fields is more and more extensive,and the navigation and positioning technology is an indispensable foundation for the normal operation of UAVs.At present,the combined method of global satellite navigation system and inertial navigation is widely used for UAV navigation.However,under the condition of satellite signal rejection,this integrated navigation method is difficult to meet the needs of UAV positioning.Therefore,the realization of the autonomous positioning algorithm of UAV under the condition of satellite signal rejection has become a hot and difficult problem at domestic and abroad.First of all,aiming at the problem that the visual/micro-inertial odometry cannot determine the carrier’s yaw in the geographic system,the combined orientation technology of micro-inertial/magnetometer/polarized camera is used to provide yaw constraints for the visual/micro-inertial odometry.Using the sliding window optimization method,a compact vision/micro-inertial combination algorithm under yaw constraints is designed.In vehicle verification experiments are carried out,and the results show that the combined vision/micro-inertial odometer with yaw constraints improves both orientation accuracy and positioning accuracy.Secondly,aiming at the divergence problem of the combined visual/micro-inertial odometry error,an autonomous UAV localization method based on semantic map assistance is proposed.A multi-level semantic feature map construction method is designed to extract the feature information and semantic information for navigation from the prior remote sensing map.A feature matching localization method assisted by odometer reconstructed point cloud was designed to estimate the position of UAV.In order to eliminate the influence of wrong feature matching on localization,a feature matching screening method based on local spatial semantic consistency was studied,and the UAV was localized using the feature matching results with the same semantic attributes in local space.The results of UAV flight experiments show that the multi-level semantic map can be used as an effective absolute constraint to provide absolute position information for UAVs,and at the same time,it can eliminate accumulated errors,thereby improving the positioning accuracy of the multi-sensor combined odometer.Finally,a UAV combined navigation algorithm with "yaw constraint + position constraint" is designed.The combined odometer under the yaw constraint can continuously estimate the navigation parameters,but the error will continue to accumulate;the localization method based on semantic map assistance does not accumulate the error,but can only achieve localization in the area with significant features.Accordingly,a combined navigation algorithm based on factor graph is designed,which fuses combined odometry under yaw constraints with the localization results assisted by semantic maps.The experimental results of UAV flight show that the designed algorithm can obtain high-precision continuous positioning results.