Research On Small Target Tracking Method Based On Unmanned Aerial Vehicle

With the development of drones and visual technology,the combination of drones with low speed and small features and target tracking has been widely concerned and applied in many fields such as urban security,investigation and investigation,and regional security,playing an important role.Due to its high flying altitude,unmanned aerial vehicles(UAVs)collect images with smaller target scales,making it difficult to accurately track.In order to solve the problem of difficulty in accurately tracking targets in unmanned aerial vehicle intelligent detection,this article will study the tracking method of unmanned aerial vehicle small targets.By establishing a model for air to ground targets,the detection,tracking,and control of air to ground targets can be achieved,and stable tracking of interested targets can be achieved.The specific research content is as follows:(1)Analysis and modeling of unmanned aerial vehicle air ground target tracking system.There is relative motion between drones and ground targets,so the flight principle and dynamic model of drones are first analyzed.Secondly,establish a drone ground to air small target tracking model to calculate the actual position and azimuth information of ground targets,thereby achieving spatial positioning of small targets.(2)Research on the detection method of small targets in unmanned aerial vehicles.In view of the influence of UAV’s own equipment and natural environment factors,which makes it difficult to detect targets,the mean filtering,histogram equalization,and optical flow methods are used to enhance image quality and reduce the difficulty of subsequent target detection.Then SE attention is introduced to optimize YOLOv5 backbone network,and on this basis,small target detection head is added,which to some extent alleviates the phenomenon of missing and false detection,thus improving the detection accuracy of small targets.(3)Research on unmanned aerial vehicle air to ground target tracking methods.This article focuses on the problem of interference and occlusion of small targets,using Siam RPN as the benchmark model and integrating it with Kalman filtering.At the same time,DIo U loss is used as the target boundary box regression to ensure the coherence of the tracking trajectory,reduce the drift of the target tracking box,and improve the accuracy and success rate of target tracking.(4)Research on unmanned aerial vehicle tracking control algorithms.Aiming at the problem of unstable flight of UAV,according to the characteristics of under actuated and strong coupling model of quadrotor UAV,a tracking controller for quadrotor UAV is designed.In order to verify the practical application ability of this design,a drone tracking suspect scene was designed.By building a drone ground target tracking platform,flight control trajectory simulation and actual video tracking experiments were conducted to verify the effectiveness and usability of the algorithm proposed in this paper.The experiments showed that the algorithm proposed in this paper can achieve the tracking of small unmanned aerial vehicle targets and pedestrians.