Research On Target Detection Method Of UAV Based On Fusion Of Infrared And Visible Images

With the rapid advancement of UAV technology and the widely use of civilian UAVs in the market,UVA has posed a serious threat to the public and individuals.To address this problem,relevant research on anti-UAV technology has received more and more attention.This topic mainly focuses on studying UAV target detection technology of anti-UAVs by making use of infrared and visible light detectors to simultaneously monitor the possibly existing UAVs in one scene for detecting,tracking and identifying UAVs in the stage of processing images.The main difficulties of this topic is to detect and track the UAV targets in different scales and to use the complementarity of infrared and visible light to detect UAV these targets synthetically.This is because the distance of the UAV target is uncertain.The main research contents are as follows:Above all,different detection algorithms are proposed for different scales of UAV targets.In response to the big target,the currently advanced convolutional neural network is used for detection,and the training carried out on the basis of the self-made data set can realize the detection towards the large target of the UAVs under different scenes and sensors.In response to small target,the traditional morphological filtering detection algorithm is studied.In order to reduce false alarms,the binary tree criterion is used to improve the TopHat transform.In addition,the ST-YOLO convolutional neural network detection algorithm for small targets is proposed based on the YOLO convolutional neural network.Performance and boundary conditions of each algorithm are also analyzed.Then,based on the detection algorithm,a tracking algorithm for multi-UAV targets with different scales is proposed.The algorithm utilizes the target tracking method of DBT(Detect before track).In the detection phase,the multi-scale target detection algorithm is run simultaneously.Based on the detection result,the trajectory management module is set up,using the trajectory association method for multi-target tracking and proposing a quadratic association algorithm combining global and local optimality.Finally,the infrared and visible image fusion recognition in the decision-level stage is carried out by using D_S evidence theory.Considering the image of a single sensor,multi-feature information of the target is extracted based on the detecting and tracking result,and the target credibility is given by using the information of different features.Meanwhile,the recognition method of the secondary fusion of infrared and visible images is proposed based on the D_S theoretical framework.The content of this method is merging the target credibility of the multi-features of the single sensor firstly,and then performing the final weighted fusion of the infrared and visible images based on size information of the target and detection information of two sensors.