Research On UAV Detection And Tracking Based On Deep Learning

An unmanned aerial vehicle,commonly referred to as a UAV,is a flight device controlled by an operator via an autonomous onboard workstation.At present,UAV has been widely used in search and rescue,security and monitoring,military communication tasks and other fields,providing great convenience for human social life.However,due to the popularity of UAVs and the imperfect control standards,a series of security problems have emerged that endanger public safety and personal privacy.Therefore,the research of anti-UAV technology becomes the key to standardize the UAV operation,and the real-time and accurate detection and tracking of UAV targets in the video is particularly important.In this paper,low-altitude UAV detection and tracking technology on the micro-edge computing platform is studied.The details are as follows:1)Aiming at the target detection problem of low altitude UAV,this paper designs a detection framework based on SSD-Mobile Net V3 network.In order to improve the accuracy of the model for low-altitude UAV target detection,a SESAM attention module is introduced based on SSDMobile Net V3 to enhance the influence of important spatial locations and channel features.At the same time,the activation function in Mobile Net V3 network is improved,so that the image features can be extracted more effectively in the low-level network,and the detection performance of UAV targets can be improved by the model.Finally,experimental verification is carried out on the self-built UAV data set.Experimental results show that the improved SSDMobile Net V3 model,compared with the original SSD-Mobile Net V3 model,although the network model parameter increases by 3M,the detection accuracy increases by 4%.Compared with the original SSD model,the number of network parameters is significantly reduced by 80%,which is more accurate for UAV target detection and more suitable for deployment on the micro edge computing platform.2)Aiming at the target tracking problem of low altitude UAV,the kernel correlation filtering tracking algorithm is used to realize target tracking.Adaptive region adjustment is introduced to solve the scaling problem of low altitude UAV targets.The problem of low altitude UAV target occlusion is solved by adding occlusion judgment and integrating Kalman filter.In order to verify the effectiveness of the improved kernel correlation filtering tracking algorithm in this paper,the self-built UAV video set is verified.The experiment shows that the improved UAV target tracking algorithm can effectively track the UAV target.3)Realize the construction and deployment of micro edge computing system platform.In parallel,the detection and tracking method is deployed on NVIDIA TX2,a micro-edge computing platform.Experiments show that the proposed algorithm can run smoothly on the micro-edge computing platform,and the speed can reach 25 fps.It can realize the real-time detection and tracking of low-altitude UAV targets,which provides certain help for the realization of counterUAV.