Research On Target Detection And Tracking Method Of Rotor UAV

With the continuous development of UAV technology and computer vision technology,the target detection and tracking technology based on UAV is widely used in our daily life.For example,in video aerial photography,drones equipped with target tracking technology can be used to automatically track and shoot aerial targets;in military operations,they can also be used to complete tasks such as enemy situation reconnaissance and enemy strikes.However,the degree of automation of early UAVs in practical applications is not high.With the continuous expansion of the demand for UAV automation work,the use of UAVs equipped with a deep learning vision platform can further realize the automation and intelligence of UAVs.I In this paper,the research on the target detection and tracking method based on the rotor UAV is carried out.The main research work is as follows:First of all,this paper summarizes the commonly used target detection and target tracking algorithms in detail,including traditional algorithms and target detection and tracking algorithms based on deep learning.The target detection and tracking method used in this paper is based on the depth of convolutional neural network.Learning method,the paper introduces some related theoretical knowledge of deep learning in detail,including convolutional layer,pooling layer,fully connected layer,and loss function and gradient optimization algorithm used in convolutional neural network,which provides the follow-up research of this paper.the theoretical basis.Secondly,in view of the current traditional target detection methods that are no longer suitable for UAV aerial work and real-time requirements,this paper studies the target detection algorithm based on the SSD and YOLO framework,and understands that the target detection algorithm based on YOLOv4 takes into account the speed and accuracy.At the same time,it is also beneficial to be deployed on the airborne computer carried by the UAV,so YOLOv4 is selected as the detection algorithm in this paper according to the application background of this paper.Then,in view of the characteristics of small pixels for the target shot from the high-altitude perspective of the UAV,this paper studies the single target tracking method based on the improved Siam RPN,introduces the basic principles of the Siam FC and Siam RPN algorithms based on the Siamese convolutional neural network,and uses the deep The network structure improves the Siam RPN algorithm.By introducing the depth residual clipping structure,the problem of inaccurate target positioning caused by the padding in the convolution operation can be effectively eliminated;and a new template update method is adopted to make it adaptable.Targets with large changes in appearance.Through the comparison of experiments on multiple data sets,the improved Siam RPN algorithm adopted in this paper has better accuracy and stability,and can achieve a relatively stable tracking effect in real scenarios.Finally,this paper builds an experimental platform based on the rotary-wing UAV,and introduces the hardware and software components of the rotary-wing UAV in detail,including the on-board computer,flight controller,image sensor,communication module and the robot used.Operating system(ROS),etc.;introduced the flight principle and control system principle of the rotor UAV.