Vision Based Ground Target Tracking Technology For Quadrotor Aircraft

Quadrotors has many advantages.For example,it has small size.It takes off and lands vertically and has excellent concealment features and so on,winning wide attention from academia and having been applied more and more widely.In recent years,computer vision technology has developed vigorously and become one of the hottest research directions,and detection and tracking technology are always research hotspots in the field of computer vision.Vision based ground target tracking system for quadrotor aircraft is a multi-functional and complex system which uses computer vision technology to detect,monitor,track and attack ground mission targets on the platform of the quadrotor.It is widely used in military and civil fields.The mathematical model of the quadrotor and a robust controller are established.The object detection algorithm based on deep learning is studied,and the correlation filter tracking algorithm is improved with the TLD framework.A physical platform is designed and built to verify the effectiveness of the algorithm.The main work of this paper is as follows:Firstly,the nonlinear mathematical model of quadrotor is established.According to the analysis of the force and moment of the quadrotor,the translation equation and the rotation equation are established.And the non-linear mathematical model of the quadrotor is derived and the aerodynamics parameters used in the simulation process are given.Then,a backstepping controller is designed based on the quadrotor model.The deduction process of the backstepping controller is elaborated in detail,and the expression of the control variables is obtained.Aiming at the disadvantage that disturbance is not considered in the backstepping controller,the disturbance observer is added to estimate the composite disturbance,and compensated the disturbance when designing the controller,also,the stability of closed-loop control system with disturbance observer is proved.The effectiveness of the designed backstepping controller based on disturbance observer is verified by detailed comparative simulation experiments.Next,considering the task of object detection,a ground target detection algorithm based on YOLOv3 and SVM is designed.The dependence of current target detection algorithms based on deep learning on large data sets is expounded,and the design process and idea of the algorithm are presented.The principles of YOLOv3 target detection algorithm,HOG feature extraction algorithm and SVM classification algorithm are explained detailedly by modules.Our own data set is made,and a large number of experiments verify the effectiveness of the target detection algorithm designed in this paper for small data sets.Following,considering the target tracking task,a target tracking algorithm based on TLD and fDSST is proposed.The framework of TLD long-term tracking algorithm is introduced,and its advantages and disadvantages are analyzed.At the same time,the virtue and shortcomings of correlation filtering algorithms are discussed.A target tracking algorithm based on TLD and fDSST is proposed to overcome the weakness of the two algorithms mentioned above.The advantages of the two kinds of algorithms are brought into full play,and the disadvantages of the two kinds of algorithms are remedied at the same time.A large number of experiments verify the effectiveness of the proposed algorithm.Finally,a qudrotor platform is built,and its hardware architecture,software architecture and target tracking scheme are introduced in detail.Based on the self-built ground target tracking system for qudrotor aircraft,flight tests were carried out to verify the rationality and effectiveness of the design of the whole platform and related algorithms.