【作者】 杨勇；

【导师】 丁勇；

【作者基本信息】 南京航空航天大学 ， 工程硕士（专业学位）， 2019， 硕士

【摘要】 基于视觉的多无人机目标追踪在无人机目标搜索、打击等方面凸显出优势,得到重要应用。本文围绕基于视觉的目标跟踪和多无人机协同避碰路径规划两个方面展开研究。针对无人机视觉目标跟踪过程中存在目标形变、光照变化、物体遮挡等问题,提出了一种改进KCF目标跟踪算法。首先,将目标的HOG特征和FAST特征进行特征融合并对其学习训练,增强了对目标表达的维度与深度,提高跟踪精确度;然后,产生虚拟目标块并用灰度直方图匹配目标,使得跟踪目标框的大小可以根据实际目标大小进行自适应变化;最后,通过仿真结果表明所提算法具有较好的跟踪精度与速度,且可实现目标框尺度随目标的自适应变化。针对多无人机协同避碰航路规划问题,提出了一种增加无人机自身斥力场的改进人工势场法(APF)与Bezier曲线相结合的方法。首先,定义了无人机自身产生的具有分段连续特点的斥力势场;其次,设计了一种基于虚拟障碍物的逃离局部最优的方法;然后,提出分段Bezier曲线平滑算法对无人机的飞行轨迹进行在线实时平滑优化,消除路径振荡现象;最后,通过仿真表明所提方法能够实现多无人机协同目标追踪与避障路径规划。针对无人机无环境模型路径规划问题,提出了一种环境信息未知连续状态下基于势函数奖赏的路径规划方法PF-DQN。首先,建立无人机在环境中的连续状态空间;其次,将360度n等分成若干个角度作为无人机的航向角,建立无人机的动作空间;接着,制定目标的势函数奖赏和障碍物的势函数奖赏;最后,通过仿真表明PF-DQN算法能够实现无人机在环境信息未知连续状态下的无环境模型路径规划,并且势函数奖赏加快了无人机路径规划的速度。设计开发了多无人机协同目标追踪实验平台。首先,介绍了实验平台的硬件设计;然后,介绍了系统软件的设计;最后,对实验平台进行了试飞实验,实验结果表明整个平台系统性、可扩展性强,是作为多无人机协同目标追踪研究可靠的算法验证平台。更多还原

【Abstract】 Vision-based UAV target tracking and obstacle avoidance highlights the advantages in the UAV target search,combat and other field,getting important application.The paper studies two aspects of vision-based target tracking and multi-UAV collaborative obstacle avoidance route planning.Aiming at the problems of target deformation,illumination change and object occlusion in the process of UAV visual target tracking,an improved KCF target tracking algorithm is proposed.Firstly,the HOG feature and the FAST feature of the target are fused.The fusion of the features is studied and trained.It enhances the dimension and depth of the target expression and improve the tracking accuracy.Then,the virtual target block is generated and the target is matched by the gray histogram,so that the size of the tracking target frame can adapt to the change of the actual target size.Finally,the simulation shows that the proposed improved KCF algorithm has good tracking accuracy and tracking speed,and can achieve adaptive change of target frame size with target size.Aiming at the problem of cooperative route planning for multiple Unmanned Aerial Vehicles(UAVs)to avoid collision,an improved artificial potential field(APF)method that combined with Bezier curve is proposed to increase the self-repulsion field of UAV.Firstly,UAV’s own repulsive potential field that has the characteristics of piecewise continuity is defined.Secondly,a method based on virtual obstacle is designed to escape from the local optimal.Then,a piecewise Bezier curve smoothing algorithm is proposed to optimize the flight path of the UAV in real time,which eliminates the path oscillation in the route planning.Finally,the simulation results show that the proposed method can achieve multi-UAV collaborative target tracking and obstacle avoidance path planning.Aiming at the problem of environment-model-free route planning for UAV,a route planning method PF-DQN based on potential function reward under unknown environmental information and continuous state is proposed.Firstly,the state space of the uav in the environment is established.Secondly,after the 360 degree is divided into several angles as the heading angle of the uav,the action space of the uav is established.Then,the potential function reward of the target and the potential function reward of the obstacle are formulated.Finally,the simulation results show that the PF-DQN algorithm can realize the environment-model-free route planning under unknown environmental information and continuous state,and the potential function reward accelerates the speed of the UAV route planning.This paper designed and developed a multi-UAV collaborative target tracking experimental platform.Firstly,it introduced the hardware design of the experimental platform.Then it elaborated the design of the system software.Finally,flight experiment is carried out.and the experiment results show that the whole platform is systematic and scalable,so it is a reliable algorithm verification platform for multi-UAV collaborative target tracking research.更多还原