Research And Implementation Of Target Detection And Tracking Of Tiny UAVs Under Complex Background Conditions

Unmanned aerial vehicle(UAV)is a kind of aircraft with small volume,light weight and strong maneuverability,widely used in civilian and military fields.The popularization of UAV promotes social progress and brings convenience to daily life.However,there are some security risks due to the disorderly flight of UAV.In addition,UAVs may be used for military reconnaissance and fire strike.Therefore,a mature anti-UAV detection technology is urgently needed to effectively supervise UAVs.In this paper,target detection and tracking of micro UAV in complex background is studied.The specific contents include:(1)A dataset of UAV in complex background is constructed.This paper proposes to obtain target datasets by data synthesis based on existing datasets,due to the lack of suitable public datasets and the high difficulty of data acquisition.First,a simple linear iterative clustering is used to segment UAV images in the La SOT dataset.Second,the trimap of the image block is obtained by dilation and erosion.Then,deep image matting model is employed to extract the UAV accurately.Finally,the multi-target UAV image is gained by data augmentation,which constitutes the target detection dataset for subsequent research.(2)An improved target detection algorithm based on YOLOv3 is proposed.In this paper,the feature pyramid structure of YOLOv3 is improved to make the network have a more detailed perception of images,so as to improve the detection performance of small targets.In addition,a residual shrinkage module is added to the basic model.This module first learns the threshold of each feature channel by attention mechanism.Then,soft thresholding is used to enhance the suppression effect of the detection model on complex background interference.(3)An improved target tracking algorithm based on DCFNet is proposed.Aiming at the boundary effect of correlation filtering,this paper proposes a coarse positioning module which can be applied to most correlation filtering algorithms.This module roughly estimates the position of the target in the search area by saliency detection based on reconstruction error.Then the search area is fine tuned to make the target as central as possible,so as to avoid the target pixels being filtered by cosine window.The improved tracking algorithm significantly enhances the tracking performance of fast motion targets.It still meets the real-time requirements,because of the high efficiency of the coarse positioning module.(4)Design of a micro UAV detection and tracking algorithm in complex background.It is mainly composed of UAV detection module,UAV tracking module,tracking evaluation module and trajectory estimation module.Among them,the detection and tracking module respectively adopt the above improved algorithm.Peak-to-Sidelobe Ratio(PSR)is used as an evaluation index of target tracking.When tracking fails,the detection module is called to initialize the tracking module.Kalman filter is the core of trajectory estimation module,which not only reduces false detection,but also improves the robustness of the algorithm when the target is occluded.The cooperative operation between modules enables the algorithm to detect UAV targets accurately and robustly,while meeting the real-time requirements.