Research On Target Detection And Tracking Algorithms In Anti-UAV Systems

The opening up of low altitude airspace has led to the booming development of the low altitude aviation market,where drones are widely used in military,commercial and civil applications due to their low cost and ease of use.The illegal use of drones and ’black flying’poses a serious threat to national security,public safety and personal privacy.For this reason,the research of anti-drone technology and derivative equipment systems has become a key topic of interest for academics and the military.Visible light detection technology has become a mainstream research method in anti-drone detection technology due to its low cost,no radiation,clear and intuitive results,and easy deployment.To address the problem of small target UAV tracking under visible light detection conditions,a research on target detection and tracking algorithms in a variety of environmental states is conducted in this paper,using multi-scale fusion and convolutional block attention mechanisms to improve the detection algorithm,and on this basis a tracking algorithm that incorporates detection models and occlusion judgment mechanisms is proposed,with the main work summarised as follows:(1)To address the problem that the original dataset collected cannot encompass all the scenes required for the experiments,this paper takes the La Sot dataset and Anti-UAV dataset as the basis,selects image sequences containing small-scale,complex backgrounds and obstacle occlusion states as the original dataset,and expands them by means of data augmentation to obtain 13,324 drone dataset suitable for the network training and experimentation in this paper.(2)As to the detection of small-scale targets,a YOLOv5 s algorithm with a four-layer detection head is proposed,which can effectively extract the feature information of small-scale targets and improve the detection accuracy of small-target UAVs;meanwhile,to address the problem of poor detection of key feature information in complex scenes.In addition,this paper adds a CBAM convolutional block attention mechanism to the detection algorithm containing a four-layer detection head to suppress irrelevant information and improve the focus on key feature information.Through experimental comparison,the detection accuracy of the P2CYOLOv5 s detection algorithm proposed in this paper reaches 94.1% on the UAV dataset.(3)A tracking algorithm based on the occlusion judgment mechanism is proposed to address the problem of obstacle occlusion in the tracking process.The relationship between the correlation quantity and the occlusion state is derived through experiments,the correlation quantity threshold is determined,and the tracking state is segmented according to the change of the correlation quantity;meanwhile,the Kalman filter is introduced for target position prediction to address the problem that the target cannot be tracked in the occlusion state.Through experimental comparison and analysis,the success rate of the improved K-KCF tracking algorithm increases by 11.8% compared with the original KCF algorithm.(4)In order to test the effectiveness of the proposed detection and tracking algorithm for UAV tracking tasks,a prototype UAV tracking system is designed and implemented based on the proposed algorithm,including three main functions of video loading,model selection and target tracking,which provides technical support for the development and practical application of subsequent Anti-UAV systems.