Research On Multi-Ground Target Matching And Cooperative Localization Method Based On Multi-UAV Vision

Sensing the location and status of ground targets through onboard visual sensors by Unmanned Aerial Vehicles(UAVs)groups is an import mission of cooperative UAVs.In recent years,UAV vision-based ground target localization has made some process and achievements.Studying the vision-based cooperative target localization by multiple UAVs is of great theoretical significance and application value.Matching sensor observation data among multiple UAVs and improving localization accuracy are two important research components of this problem.However,due to the low resolution of image and blurred target features,limited onboard sensor accuracy,and unsatisfactory communication,the accuracy of target matching and cooperative localization needs to be further improved.The main work and contributions are as follows.(1)The error distribution and optimal observation configuration of multi-UAV target fusion localization are analyzed.A Monte Carlo simulation is processed to analysis the localization accuracy based on the angular rendezvous positioning model.The relative position relationship between UAV and target has a significant influence on the positioning accuracy.The optimal observation configuration of UAVs is obtained.(2)A multi-view multi-target matching method with restricted image features in the UAV viewpoint is proposed.For target detection for small targets,a lightweight target detection algorithm(UAV-YOLOv4-tiny)is proposed.For the target matching under the conditions of restricted image features,the siamese neural network target matching method based on the Central-Surround dual data stream is proposed.Based on the Deep SORT target tracking algorithm,image position over a longer time range is achieved.Finally,the open-source and self-built datasets are used to verify that the proposed algorithm,the matching accuracy is 96.42%.(3)The end-to-end target cooperative localization and state estimation algorithm is proposed to improve the target localization accuracy.For the observation noise and measurement error of airborne sensors,the Unscented Kalman Filter(UKF)is used.For the random motion of non-cooperative targets,the Interacting Multiple Model(IMM)algorithm is used.Considering the asynchronous measurement caused by communication delays and pre-processing times,the proposed algorithm is improved for asynchronous measurement and asynchronous non-sequential measurement.Finally,simulation experiments are used to verify that the algorithm proposed can effectively reduce the target localization error compared with other cooperative target localization algorithms by at least 32.16%.(4)A multi-UAV cooperative localization flight environment is designed to verify the effectiveness of the proposed algorithm under realistic scenarios.Three multi-rotor UAVs and ground target vehicles are used.Experiments are processed in urban and field environments.The effectiveness of the proposed end-to-end localization algorithm is verified by comparing with other multi-UAV cooperative target localization algorithms in real scenarios.The results show that the proposed method reduces the localization error by at least 15.41% in the 3D direction.And the time-asynchronous end-to-end localization algorithm reduces the localization error by 12.75% compared to the synchronous end-to-end localization in real scenarios.