Research On Autonomous Intelligent Monitoring Algorithm Based On UAV

In recent years,with the continuous maturity of drone control technology,drones have been widely used in various fields.As an important security tool in modern social life,video surveillance seriously affects public safety and order.Existing video surveillance algorithms are mainly based on fixed-point cameras,which have the disadvantages of high cost and single viewing angle.In response to this problem,this paper proposes a research on autonomous intelligent monitoring algorithms based on UAVs.The main research work and innovation are summarized as follows:First,an overall architecture of the UAV intelligent monitoring algorithm is designed.In view of the unclear definition and unscientific architecture of the current UAV intelligent monitoring,the overall architecture of the UAV intelligent monitoring algorithm is proposed.The algorithm is divided into the UAV target detection and tracking algorithm that realizes the intelligent monitoring function and the UAV autonomous landing algorithm that improves the UAV’s life time.And in-depth analysis of the research difficulties of intelligent surveillance algorithms from the perspective of UAVs,to provide motivation and theoretical basis for the algorithm proposed and improved in the subsequent chapters.Second,based on the existing target detection and tracking algorithm,in view of the current UAV viewing angle,the target size is small and the background is complex,and a target detection and tracking algorithm suitable for UAV platforms is proposed.Specifically,the data augmentation method and training strategy of the target detection algorithm YOLO V4 are improved,so that the algorithm’s m AP index on the UAV target detection data set Visdrone2019-DET is increased from 26.73 to 29.72.The positive and negative sample division method and loss function of the target tracking algorithm Siam FC++ are improved,so that the AUC index of the algorithm on the UAV target tracking data set UAV123 is increased from 57.4% to 60.2%.Experiments on the UAV target detection and tracking algorithm are carried out using the data set in the real scene.The results show that the algorithm also has a good visual effect in the real scene and has good generalization.Third,a vision-assisted autonomous landing algorithm for UAVs is proposed.Specifically,design general stopping signs,make stop sign data sets Sign1 and Sign2,and propose a multi-stage accurate identification and positioning algorithm based on stop signs to assist autonomous landing.The algorithm designed a two-level recognition and positioning algorithm based on the height information of the drone.The first-level recognition and positioning algorithm uses a deep learning detection algorithm.In order to improve the accuracy and generalization of the detection algorithm,the data set is expanded by synthetic data.The m AP of the first-level recognition and positioning algorithm on the data set Sign1 proposed in this paper reaches 72.68;the second-level recognition and positioning algorithm uses the graphic information of the stop sign to identify the position offset and the angle offset of the stop sign to realize the drone’s Adjust the angle to improve the landing accuracy.The second-level recognition and positioning algorithm has an accuracy of 88.5% on the data set Sign2 proposed in this paper.Fourth,an autonomous drone monitoring system is built.In view of the high cost and low performance of the current UAV onboard processor,an autonomous UAV monitoring system has been built,which uses a back-end processing server to process surveillance video data.The system specifically includes six components: drone,remote control,streaming media server,back-end processing server and management client.Use video push-pull streaming technology to achieve low-latency,high-quality video data transmission.Use UDP,TCP technology to realize the communication between the management client and the back-end processing server.At the same time,in order to facilitate the operation and debugging of the staff,a simple operation,generous and concise operation interface is designed.