Research On Robust Tracking Method Based On Human-machine Collaboration For UAV

Unmanned Aerial Vehicle(UAV)can conduct multi-angle,large-scale,and long-term observation of ground targets without being restricted by terrain,so as to obtain comprehensive and rich mission information.However,these features of UAVs also bring more challenges to visual target tracking.Most current visual tracking algorithms are unable to cope with the rapid changes in the appearance of targets in complex scenes and the long-term disappearance of targets,which makes it difficult to achieve robust target tracking in complex environments from the perspective of drones.In order to achieve robust visual target tracking on the UAV platform,this paper combines the UAV’s "human-in-the-loop" supervisory control method,and proposes a human-machine hybrid target tracking method based on eye-movement interaction technology.Human cognitive intelligence is introduced into the tracking loop of the machine through eye movement interaction technology,thereby improving the robustness,accuracy and practicability of the target tracking method.The main content of this article includes innovative points:(1)Aiming at the interactive tasks in the scene of the drone ground control station,a fast target selection method based on gaze control is designed for the static and dynamic targets in the interactive interface,and the static virtual keyboard is used as the research object to verify.Experimental results show that the target selection method based on secondary triggering of eye gestures can greatly reduce the possibility of "Midas touch" and maintain good input efficiency,which is suitable for tasks that require long-term interaction.(2)In response to the long-term tracking challenges of UAVs,a hybrid long-term target tracking framework is constructed.The framework reasonably allocates the functional authority of humans and machines in the UAV long-term tracking task,fully guarantees the human intervention and decision-making power,and maximizes the machine’s advantages in large-scale data processing without fatigue.Through the analysis of human attention,a dynamically updated feature library of human interest targets is constructed,which enhances the machine’s understanding of tracking targets.When the target disappears for a long time and reappears or the appearance changes sharply,the target can be retrieved accurately.In order to verify that the tracking framework is more effective in the complex environment of UAV tracking,this paper also constructed a pedestrian target test data set of UAV in the complex environment.Experimental results show that the tracking method proposed in this paper greatly exceeds the performance of advanced algorithms such as SPLT,ATOM,and Da Siam RPN on this test set,indicating that the method proposed in this paper has a stronger tracking ability in the complex environment of UAV pedestrian tracking.At the same time,this article explores the impact of human attention on the framework.The results of ablation experiments show that human participation has a positive effect on the tracking performance of the tracking framework,and the more human attention to the target,the more it can enhance the tracking performance of the machine.(3)On the basis of the first two studies,this paper designs and implements a pedestrian-following prototype system of UAV with gaze-assisted control.First of all,a method of gaze-assisted drone control is designed according to the eye movement characteristics of the operators of the drone ground station.Subsequently,the human-machine hybrid visual tracking algorithm is used as the image processing method in the system to obtain the target position in the image,and the UAV is visually servoed to control the UAV to follow the pedestrian target.Experimental results show that the prototype system can effectively complete pedestrian following tasks,and can cope with complex changes such as target disappearance and appearance changes,and realizes the robust autonomous follow-up of pedestrian targets by the UAV.