Research On Path Planning And Target Recognition Technology Of Unmanned Aerial Vehicles

Unmanned Aerial Vehicles(UAV)is gradually integrated into the public life,and gradually becomes the focus of attention and research due to its small size,simple structure and wide use.Especially in the exploration field of UAV,UAV has more and more applications,such as,it is easy for UAV to reach the designated location for target detection with camera and lidar.Moreover,UAV has its own advantages,such as simple mechanical structure and control theory,low manufacturing cost and so on,which is conducive to the promotion in various fields.In this paper,the control method,path planning and obstacle identification of the aircraft are designed and completed.Firstly,this paper introduces the research summary of UAV at home and abroad,summarizes the research progress of domestic and foreign scholars,and describes the path planning algorithm(Rapidly Exploring Random Tree,RRT)and target recognition algorithm in detail.Secondly,in this paper,the model of UAV is introduced including altitude control,Roll,Pitch and Yaw control,and the current popular dynamic modeling of aircraft is introduced including Euler method and quaternion method,as well as the transformation relationship between the two methods.Thirdly,the basic principle and parameter setting of RRT path planning algorithm are introduced,studies the influence of different step size and collision radius on the running time and optimal path of the algorithm,obtains the optimal path in twodimensional space and three-dimensional space.Fourthly,when the process of path planning is finished,the planning trajectory needs to be smoothed to meet the flight requirements.This paper introduces the application of gradient descent method to meet this qualification.Finally,this paper identifies the car model by YOLO V3 the deep neural network algorithm.this algorithm adopts the Darknet-53 network architecture,draws lessons from the practice of residual network,sets fast links between some layers,forms a deeper network level,and multi-scale detection,which improves the recognition accuracy and the detection effect of small objects.Compared with the current research data in the field of UAV,the experimental results show that the recognition accuracy is similar,and the mission of recognizing obstacle car can be finished.