Research On Urban Low-altitude Multi-rotor UAV Image Recognition And Tracking Method

In recent years,civilian multi-rotor UAVs are widely used in various fields,bringing convenience to people at the same time also producing a lot of negative impacts,the phenomenon of UAV black flight has affected people’s normal life.For critical places such as governments,hospitals,universities,and research institutes,there is an urgent need to establish efficient countermeasures against UAVs.In this regard,this paper improves on the existing target detection algorithms,proposes multi-rotor UAV detection methods in static and dynamic backgrounds,respectively,and deploys them to UAV interception systems.The main works of this paper are as follows.(1)This paper proposes a UAV detection method based on improved frame difference and classification networks.Firstly,the algorithm uses the frame difference to output the minimum bounding box of foreground pixels,dynamically calculates the connectivity distance between the bounding boxes by the size of each bounding box,fuses Breadth First Search to cluster the bounding boxes,and generates the initial detection box,then uses Res Net50 embedded with Efficient Channel Attention to classify and filter the detection frames.Experiments show that the algorithm in this paper can achieve fast target detection in complex environments,and the detection speed can reach 27.58 fps.Compared with Res Net50,the precision,and recall are improved by 3.35% and 6.51%,which meet the accuracy and real-time requirements in monitoring scenarios.(2)This paper proposes a recognition and tracking method for UAVs in complex environments.The first step is to use YOLOv5 for identifying UAVs.Then we combine the motion trajectories of UAVs in continuous video frames to correct the error results.Finally,using kernel correlation filter(KCF)to improve the ability to continuously track the target UAVs.Experiments show that this method can effectively improve the recognition rate.Compared with the YOLOv5 s network,the precision has improved by 9.13%,the recall has improved by 11.54%,and the recognition speed can reach to 27.64 fps.After testing the proposed algorithm on the self-developed UAV interception platform,the results show that it can continuously track targets and efficiently intercept the invading UAVs.(3)This paper proposes an optimization and improvement method for a multi-rotor UAV interception system.This approach starts by simplifying the interception system using an embedded edge computing device,deploying an improved YOLOv5 algorithm,and achieving fast detection for a single frame of less than 25 ms.Then,it connects the interception system with multiple surveillance cameras through the network to achieve stable transmission of data from a wide range of surveillance targets.Finally,the supporting Android monitoring software is designed and developed to realize the expected functions such as environment monitoring and intrusion warning.