【作者】 张曦；

【导师】 郑荣濠；

【作者基本信息】 浙江大学 ， 电气工程， 2018， 硕士

【摘要】 近年来,随着四旋翼无人机的载重能力、续航能力、安全性和灵活性的大幅提高,利用四旋翼搭载摄像头进行目标搜索或跟踪成为新的研究热点。目前已有研究能够实现对单个目标的识别或跟踪,然而在某些应用场合,例如足球、赛车、自行车等室外体育赛事直播中,往往无法也无需对每个目标配置单独的跟踪设备,此时就需要四旋翼同时跟踪多个运动目标。同时,实际应用中各个目标的重要性是不同的,例如在体育赛事中执行跟踪任务的四旋翼应该适当向明星运动员靠拢,这就要求对四旋翼的跟踪路径进行优化。为实现以上功能,本文提出了一种基于k-means聚类算法和加权多目标重心跟踪的新型算法,能够控制四旋翼识别多个目标,并根据目标的特征信息对不同目标赋予不同权值,然后根据各个目标的位置和权值获得跟踪路径,并控制四旋翼沿此路径进行多目标跟踪。然而在一些情况下,比如空间探测,恶劣环境搜救等诸多应用场景,由于单架四旋翼自身功能的限制难以满足复杂任务应用需求,通常要采用多架四旋翼协作完成任务,多架四旋翼协同控制也已成为无人机研究的一个非常重要的研究方向。因此在本文采用Odroid/Intel NUC机载电脑和DJI M100飞控搭建了多四旋翼协作的实验平台,组建了多机通信的无线Mesh网络,使得多四旋翼之间能够通过数据通信感知相邻四旋翼的信息,实现了基于ROS的分层的软件系统架构设计,完成了多飞行器的分布式三角编队算法。最后提出了两种多机协同目标搜索算法,一种是基于缩放型三角编队的多机目标搜索,另一种是基于信息融合的分区域协同搜索。前者是一种基于遍历思想的固定路径搜索算法,后者是一种基于动态路径优化的搜索算法。本文给出了详细的理论阐述和仿真结果。本文的创新点和贡献在于以下四点:第一,提出了一种基于k-means聚类算法和加权多目标重心跟踪的新型算法,该算法利用搭载的单架四旋翼平台实现多目标的搜索和动态跟踪;第二,搭建了基于socket通信的多机编队硬件平台,并完成了基于ROS的高可移植性的跨平台软件系统;第三,提出了一种基于缩放型三角编队的多机联合的目标搜索算法,将多机编队和目标搜索相结合,并通过仿真验证了该算法的可行性;第四,提出了一种基于信息融合和协同控制的多机分区域搜索算法。本文给出了理论分析和仿真结果。更多还原

【Abstract】 In recent years,with the quadrotor load capacity,endurance,safety and flexibility greatly improved,the use of quadrotors equipped with cameras for target seeking or tracking has become a new research hot spot.However,in some applications,such as soccer,racing,cycling and other outdoor sports events,it is often impossible to configure a tracking device for each target.In this case,a quadrotor can be used to track multiple moving targets at the same time.Meanwhile,the importance of each goal in practical application is different.For example,the quadrotor performing the tracking task in a sporting event should be appropriately moved closer to the star athlete.This requires optimization of the tracking path of the quadrotor.In order to achieve the above functions,a new algorithm based on k-means clustering algorithm and weighted multi-target gravity center tracking is proposed in this article,which can control the quadrotor to recognize multiple targets and assign different weights to different targets according to the target characteristic information.Then the tracking path can be obtained according to the position and weight of each target,and the quadrotor is controlled along the path for multi-target tracking.However,in some cases,such as space exploration,rescue in harsh environment search,due to the limitation of single-quadrotor’s own functions,it is difficult to meet the needs of complex mission applications.Usually,more than one quadrotor is used to accomplish the task.Multi-quadrotor cooperative control has become a very important research direction of UAV research.Therefore,in this article,a multi-quadrotor collaborative hardware experiment platform is built using Odroid/Intel NUC airborne computers and DJI M100 flight controller.Based on this,a wireless Mesh network with multi-quadrotor communication is set up,which enables multi-quadrotor to obtain the information of the neighboring quadrotor via data communication.The layered software design system based on ROS is completed and the distributed delta formation of multi-quadrotor system is realized.In the end,two kinds of multi-quadrotor collaborative target search algorithms are proposed.one is multi-target search based on scaled triangular formation and the other is sub-regional collaborative search based on information fusion.The former is a traversal fixed-path search algorithm,the latter is based on dynamic path optimization search algorithm.This article gives a detailed theoretical analysis and simulation results.The contributions of this article lie in the following four points:Firstly,a new algorithm based on k-means clustering algorithm and weighted multi-target gravity center tracking is proposed.The algorithm is based on a single quadrotor to achieve multi-objective search and dynamic tracking;Secondly,a multi-quadrotor formation hardware platform based on socket communication was built.Meanwhile a cross-platform software system with high scalability and high portability based on ROS was completed;Thirdly,this article proposes a multi-quadrotor joint target search algorithm based on zoom-type delta formation,which combines the multi-quadrotor formation with target search,and the feasibility of the algorithm is verified by simulation;Fourthly,A multi-quadrotor sub-region search algorithm is proposed,which is based on information fusion and target search under collaborative control.This article gives theoretical analysis and simulation results.更多还原