Research On Target Tracking Algorithm Of Tethered UAV Based On Siamese Network

In recent years,with the rapid development of technology,an increasing number of specialized unmanned aerial vehicles(UAVs)have appeared in various fields such as agriculture mapping,search and rescue,border patrol,and military reconnaissance.Tethered UAVs are a type of UAV connected to a ground station through a cable or tether,allowing them to perform long-duration high-altitude flight tasks.However,tethered UAVs are limited by the length of the tether and typically used for emergency rescue scene illumination tasks.This study focuses on tethered UAVs with the goal of enabling target tracking functionality.The research emphasizes the development of lightweight target tracking algorithms suitable for UAV platforms and the design and construction of a tethered UAV target tracking system.The specific research work is as follows:(1)To address the limitations of traditional Siamese target tracking algorithms,which are bulky and difficult to deploy on embedded devices,and their tracking performance in conditions with large scale changes or interference from similar objects,we propose a new lightweight fast tracking algorithm called Ghost Fast Tracking with Ti FPN and Retriever(GTtracker).This algorithm introduces the Ghost mechanism to redesign the Resnet network,creating a lightweight G-Resnet network for rapid target feature extraction.It then employs a lightweight adaptive weighted fusion algorithm,Tiny Adaptive Weighted Fusion Algorithm Feature Pyramid Network(Ti FPN),to enhance feature information fusion and address interference from similar objects.A lightweight region regression network,Ghost Decoupled Net(GDnet),is designed for target classification,IOU calculation,and bounding box regression.A novel target retriever is applied during the tracking phase to improve the success rate of the algorithm.The algorithm is validated on the OTB100 and VOT2020 datasets and performance-tested on the embedded device Jetson Xavier NX.(2)To tackle the issue of the declining success rate in long-term tracking tasks due to accumulating errors and difficulty in target retrieval when the target is temporarily out of the UAV’s field of view,we first introduce common target template update strategies and analyze the causes of tracking errors in existing algorithms.We then propose an online template update mechanism based on Meta-Master to address accumulated errors and target retrieval issues in long-term tracking tasks.This mechanism is integrated into the GTtracker algorithm framework to form the lightweight long-term tracking algorithm MGTracker,which is validated on the VOT2018 LT and La SOT datasets.Based on the aforementioned algorithms,we constructed a tethered UAV platform with a car as the primary tracking target.The algorithm was deployed on an onboard computer for UAV flight control.The UAV captures and identifies the car’s movement through its camera and performs tracking flight tasks.When the tracking target appears in the UAV’s camera and is identified by the onboard computer,the target tracking algorithm and flight control algorithm are used to control the UAV’s movement,reducing the distance between them while maintaining a safety threshold.If the distance is less than the safety threshold,the UAV will decelerate or hover.In this study,we developed a tethered UAV system,which runs the target tracking algorithm on an embedded device,NVIDIA Jetson NX,to control the UAV’s flight controller and achieve target tracking functionality.The effectiveness of the lightweight target tracking algorithm was validated through Gazebo simulation experiments,and the feasibility of the proposed algorithms was confirmed through experiments on the tethered UAV platform.This research demonstrates the practical significance and potential application value of our work.