Research On UAV Tracking System Based On Machine Vision

Autonomous flying drones have become important tools not only in the military field but also in the civilian field.The global drone market is currently hot.The market value of the drone in 2019 predicted by market analysts five years ago is about 63.50 100 million euros.And now,the new assessment shows that the global drone market is worth 15.42 billion euros and will grow to 44.46 billion euros by 2025.It can be seen that the global market of drones is developing faster than expected.At the same time,the field of UAV tracking is an important part of the practical application of UAVs.UAV autonomous tracking of ground moving objects has recently become an important research topic in the public literature.In the military field,UAV tracking can be used for the reconnaissance and strike of mobile military targets on the ground,after the UAV recognizes and discovers the target through various airborne sensors,it conducts stable tracking and reconnaissance while maintaining a certain safety distance and is not found.Compared with the traditional dispatch of reconnaissance vehicles and the tracking method of personnel,UAV tracking has the advantages of visual field and terrain,and is also more secure.In the civil field,drone autonomous tracking can be used for the pursuit of pedestrians in traffic accidents.Compared with the traditional dispatch of human or vehicles to track the perpetrators and vehicles,the UAV platform is less affected by the terrain and has a wider field of vision.This paper designs a complete set of UAV visual tracking control system.First,the core tracking algorithm of UAV target tracking is improved to make the algorithm more robust against occlusion,scale changes and background confusion;Secondly,the intelligent network connection and embedded algorithm are designed to transmit the images captured by the drone in the air to the ground in real time and implement the detection algorithm at the ground station to transmit the results back to the drone control terminal,so that the operation can be broken The airborne processor has limited computing power and implements a more accurate algorithm;next,the hardware is used to build the UAV physical platform;finally,the simulation and physical test are designed separately to realize the system tracking in simulation and actual environment Flight,analysis and demonstration of the feasibility of the system described in this article.The main research contents of this article are as follows:1.Visual perception algorithm part: First,this paper proposes a multi-scale kernel correlation with adaptive learning rate for the problems of scale changes,background chaos,tracking drift and tracking failure caused by visual target tracking of drones.The filtering tracking method has been proved by experiments that the algorithm proposed in this paper can track the target more accurately when facing the above problems.Second,the relative position of the drone and the target to be tracked by the drone multisensor fusion algorithm is used in this paper.Posture solution.The airborne ultrasonic sensor,acceleration sensor and GPS are used to obtain the UAV’s own pose,and the relative position of the UAV and the ground target is obtained after the fusion of the coordinate transformation and the visual perception results.2.Intelligent network connection part: This article designs a set of UAV intelligent network connection system.Nowadays,4G and 5G network capabilities meet the communication requirements of most application scenarios of drones.This article uses the Io T protocol MQTT The development has completed the low-latency cloud upload and backhaul of the high-definition image of the drone,the cloud upload and backhaul of the sensing results of the drone’s airborne sensors,and the ground station system performing low-time on the air drone through the Io T protocol Extend control.The ground-end control of the UAV take-off is realized,and the tracking target is selected for stable and autonomous tracking by the UAV platform.