Research On Aerial Target Recognition And Tracking Technology

With the rapid development of deep learning technology,target recognition and tracking technology has gradually become a research focus in the field of computer vision.Whether in military or civilian use,the recognition and tracking of aerial targets is an important application of recognition and tracking technology.As a common target in the air,small UAVs have the characteristics of small size,fast flight speed,strong maneuverability,and complicated flight environment,which have caused great difficulties in image and video-based aerial target recognition and tracking.In view of this,this article studies the method of target recognition and tracking for small drones in videos.The research contents include data acquisition and image preprocessing,construction of data sets,target recognition of drones,and target tracking of drones.The main work of this article is as follows:(1)Collect video data with small UAVs as aerial targets Firstly.Organize common image preprocessing methods,write video processing programs,and process the collected video data into fixed-format image data.These image data are then annotated manually to build a small UAV data set.(2)After analyzing the data set,it is found that the target size of the drone is too small and has certain special characteristics.Aiming at this problem,this paper proposes a re-clustering target recognition(RCTR)method for small drones based on a deep neural network and a target recognition model YOLOv3.This method solves the problem that traditional algorithms have low recognition accuracy for small-sized UAVs.(3)On the basis of identifying the drone,further research on the tracking problem of the drone.In this paper,K-nearest neighbors are used to propose a multi-feature fusion tracking(MFFT)method.This method adds position prediction,improves the accuracy and robustness of tracking,and solves the problem of easy tracking loss of aerial drones.