| In complex environment missions such as aerial reconnaissance and fugitive pursuit,post-disaster search and rescue,UAVs are receiving increasingly widespread attention for their agile motion capabilities,and one of the key issues in accomplishing these missions is target tracking.Common features of complex environments in such missions include the absence of a priori information about the environment before conducting the mission,the presence of unknown dense obstacles in space,the absence of GNSS satellite signals can only rely on airborne equipment to complete navigation,and target detection will be affected by obstacle occlusion.UAV tracking of moving targets in such environments requires comprehensive consideration of target detection and positioning and motion planning in complex environments.To prevent target loss and ensure the tracking effect,it is also necessary to introduce multiple UAVs for joint tracking.Therefore,how to make UAVs capable of both autonomous obstacle avoidance and continuous tracking of maneuvering targets in complex environments and how to collaborate the tracking tasks among multiple UAVs are issues that need to be focused on.To this end,this paper investigates technologies related to UAV tracking of dynamic targets in complex environments,including:1.The problem of locating and predicting the trajectory of moving targets in complex environments is studied.Firstly,an autonomous UAV navigation scheme is constructed to complete the UAV’s own localization and map building in the complex environment.Then an improved YOLO algorithm is proposed for detecting motion targets of unknown intent by using on-board vision,which can improve the robustness of the tracking system while effectively avoiding false detection by fusing human pose in the YOLO detection result frame to achieve tandem detection.Then,for the target trajectory prediction problem,an improved Bessel curve target trajectory prediction algorithm incorporating human pose is proposed.The target escape behavior prediction algorithm is designed by the pose relationship of human joints,which can effectively predict the target motion intention and ensure the continuous tracking of the target.The experimental results show that the improved YOLO algorithm reduces the false detection rate by 10% and the missed detection rate by 5% compared with the single YOLO detection method.In terms of trajectory prediction,compared with the traditional Bessel curve prediction method,the trajectory prediction algorithm in this paper improves the prediction accuracy by 16.2%,reduces the prediction calculation time by 13.6%.2.The problem of UAV trajectory generation in complex environments is studied.A hybrid A* path search algorithm considering the predicted future motion of the target is designed at the front end to generate feasible tracking paths.Then,an unconstrained minimum control volume trajectory representation is used in the back-end to optimize the spatio-temporal trajectory based on the feasible path in the front-end.And by laying down a safe flight corridor to separate the UAV from the surrounding obstacles,the redundant obstacle avoidance constraints in the trajectory optimization process are eliminated,thus improving the speed of trajectory planning.In the trajectory optimization part,the visual tracking task constraint is designed considering the obstacle occlusion effect to ensure the observation distance and observation angle,which can ensure the safety while effectively tracking the target.Finally,the effectiveness and superiority of the trajectory generation and optimization method in this paper are verified by comparison experiments.The experimental results show that compared with the existing literature,the effective tracking time of this paper’s tracking method in the same environment is improved by more than18% and the computation time is reduced by more than 15%.3.For the problems of limited field of view and insufficient detection range of single UAV tracking system,based on the above-mentioned single-aircraft system with high autonomy,the single-aircraft system is further extended to a multi-aircraft tracking dynamic target system.We focus on the multi-UAV cooperative target tracking algorithm and design the inter-aircraft collision avoidance and positioning drift compensation scheme,which can effectively solve the problem of limited field of view and detection range of single UAV through target information sharing and reduce the probability of target loss.The probability of target loss is reduced by target information sharing.Finally,the above algorithms are verified by real aircraft flight tracking experiments in complex environments.The experimental results show that the multi-aircraft system achieves efficient target tracking,which proves the effectiveness of this paper’s algorithm and expands the application prospect of this paper’s method for multi-UAV cooperation. |
PDF Full Text Request