Deep Learning Analysis Of Group Based On UAV

With the improvement of the intelligent level of drones,drones are more and more widely used in military confrontation and our life,especially the aerial photography of drones provides great convenience for data collection.The multi-vision data acquired by drone cluster is greatly improved compared with the monocular data of the single drone.In order to achieve real-time sampling of the number of people in different places,this paper builds a real-time,high-precision and distributed heterogeneous drone crowd analysis system.The real-time system ensures instant and efficient communication within the drone cluster and the real-time data.High-precision demographic algorithms provide accurate population number for people surveys.Distributed heterogeneous features ensure that each drone performs different control strategies,ensuring high-speed flight and security of the cluster in the mission.This paper is mainly based on statistical analysis of existing populations,using drone aerial photography as a data carrier to design and implement a real-time,high-precision and distributed heterogeneous drone crowd analysis system.The main research contents are as follows:1.According to the real-time,high-precision and distributed heterogeneity of the system,based on the movement of the autonomous drone cluster in the closed space,an experimental scheme for open space drone cluster control and autonomous obstacle avoidance is proposed.The scheme adopts the unmanned aerial vehicle sensor to detect the surrounding environment,and completes the function of autonomous obstacle avoidance of the drone cluster with limited information;The distance between the drone and the obstacle is used as a reference parameter to make the drone pace of obstacle avoidance smoother,and the stability of the drone observation is improved with a smaller speed change.2.This paper completed the optimization of the demographic framework based on the C^3 Framework,used fast head calibration to accurately locate the number of people in the picture,optimized the framework for the problem of inaccurate head frame and improve the accuracy of the number of people's statistical estimates,made self-sampling,tuning,and training models further improve the accuracy and robustness of population density prediction.3.This paper completed the real machine test and verification of drone cluster control and aerial photography.The test for the flight performance of the UAV cluster,tests the flight,obstacle avoidance,and the execution of the aerial mission and the return data picture,which can provide clear aerial data for the follow-up statistics.And through the analysis of the number of people,it is predicted to obtain high-accuracy scene number data.The test results show that the standard deviation of the UAV flight speed of the system at the specified cluster speed during the obstacle avoidance is reduced by 50% to 80% for the traditional artificial potential field.The drone has a smaller speed deflection when the obstacle is avoided,and the stability is greatly improved.In terms of the number of people in the test environment,the accuracy and stability of the algorithm are improved.