【作者】 王红；

【导师】 彭宇；

【作者基本信息】 哈尔滨工业大学 ， 仪器科学与技术， 2021， 硕士

【摘要】 虽然无人机以其低成本、高安全性以及强机动性等优点具有广泛的应用,但是其对外界环境及人员会带来潜在的安全威胁。因此,有必要开展无人机的目标检测与跟踪方法研究,为开展有效地无人机飞行监控提供技术支撑,达到避免发生人身伤害和物质损伤。无人机的目标检测根据其检测设备的不同主要分为机器视觉和激光雷达两类方法。其中,机器视觉的方式是采用相机等视觉传感器,借助于计算机视觉领域的方法与技术研究实现目标的检测。因为视觉传感器成本低以及易于产业化,所以机器视觉的方法广泛应用在视频监控和无人驾驶等领域。然而,机器视觉的方法较难探测到目标的深度信息,对于无人机的检测以及跟踪而言,影响其实际应用效果。相比于机器视觉的方法,使用激光雷达作为传感器能够探测到目标的深度信息,因此本文采用基于激光雷达的方法实现无人机的检测和跟踪。本文具体的研究内容如下:首先,将目前应用较为广泛的旋翼无人机作为研究对象,提出基于激光雷达的目标检测与跟踪的总体方案。在硬件方案设计方面,深入分析国内外典型产品的特性后,确定适用于本文的混合固态型激光雷达,同时结合目前多类型嵌入式处理器特征和数据处理需求,选择异构片上系统Zynq实现数据采集与处理;在软件方案方面,设计数据采集程序所需的交叉编译方案,提出适用于点云数据的聚类检测算法和基于粒子滤波的跟踪算法方案。其次,在软件算法方案的基础上,提出基于欧式距离聚类的目标检测算法和基于粒子滤波的目标跟踪算法。无人机点云数据样本的稀疏性导致较难应用机器学习的方法进行目标检测,因此本文采用基于欧式距离聚类的算法进行对目标的检测,通过设置聚类中一个类别容许的最大点云数和最小点云数以及相邻两点之间的容忍度阈值实现无人机的检测。在此基础上,利用粒子滤波算法可以解决非线性非高斯问题的优点,构建无人机的状态转移方程,同时建立激光雷达的观测方程,通过建立粒子滤波的模型实现无人机的跟踪。利用基于聚类和粒子滤波的方法,完成对无人机的检测和跟踪。最后,完成无人机检测与跟踪的综合实验验证。利用异构片上系统实现对激光雷达的数据采集单元的构建,完成Linux操作系统的移植,搭建可以运行算法的操作环境。在操作系统的环境下,利用交叉编译工具在主机上生成可以在异构片上系统运行的可执行程序,实现数据采集程序的移植。为验证算法的有效性以及硬件功能的实现,设计并完成三组实验,其中检测精度实验验证基于聚类的目标检测算法的准确性,检测范围实验表明激光雷达能够检测到无人机的范围,无人机动态飞行实验验证粒子滤波算法可以实现对无人机的跟踪,同时能够验证Linux操作系统的成功移植以及数据采集程序的正确运行。更多还原

【Abstract】 Although Unmanned Aerial Vehicle(UAV)have a wide range of applications due to their low cost,high safety,and strong maneuverability,they will pose potential safety threats to the external environment and personnel.Therefore,it is necessary to carry out research on UAV target detection and tracking methods to provide technical support for effective UAV flight monitoring to avoid personal injury and material damage.The target detection methods of UAV are mainly divided into two types,including machine vision and lidar according to the different detection equipment.Among them,the method of machine vision is to use vision sensors such as cameras to achieve target detection with the help of methods and technology research in the field of computer vision.Due to the low cost and easy industrialization of vision sensors,machine vision methods are widely used in fields such as video surveillance and unmanned driving.However,it is difficult for the machine vision method to detect the depth information of the target,which affects its practical application effect for the detection and tracking of the UAV..Compared with the method of machine vision,the use of lidar as a sensor can detect the depth information of the target,so this article adopts the method based on lidar to realize the detection and tracking of UAV.The specific research content of this thesis is as follows.Firstly,the widely used rotary-wing UAV is taken as the research object.An overall framework based on lidar for target detection and tracking is proposed.In terms of hardware design,after in-depth analysis of the characteristics of typical products at home and abroad,a hybrid solid-state lidar suitable for this article was determined.At the same time,combined with the current multi-type embedded processor characteristics and data processing requirements,the heterogeneous system on chip Zynq is selected to realize the data acquisition and processing.In terms of software schemes,the cross-compilation scheme required by the data acquisition program is designed,and a clustering detection algorithm suitable for point cloud data and a tracking algorithm scheme based on particle filtering is proposed.Then,Based on the software algorithm scheme,a target detection algorithm based on Euclidean distance clustering and a target tracking algorithm based on particle filter are proposed.The sparseness of the UAV point cloud data samples makes it impossible to apply machine learning methods for target detection.Therefore,this thesis uses an algorithm based on Euclidean distance clustering to detect the target,by setting the maximum number of point clouds allowed in a category in the cluster.And the minimum number of point clouds and the tolerance threshold between two adjacent points realize the detection of UAV.On this basis,the particle filter algorithm can solve the advantages of non-linear and non-Gaussian problems,construct the state transition equation of the UAV,and establish the observation equation of the lidar at the same time,and realize the tracking of the UAV by establishing the particle filter model.Using methods based on clustering and particle filtering,the detection and tracking of UAVs are completed.Finally,comprehensive experiment evaluations of UAV detection and tracking are implemented.The data acquisition unit of LiDAR is constructed by using the heterogeneous system on chip.It mainly completes the transplantation of Linux operating system to build an operating environment that can run algorithms.In the operating system environment,cross-compilation tools are used to generate executable programs that can run on the heterogeneous system on chip on the host computer,so as to realize crosscompilation of data acquisition programs.In order to verify the effectiveness of the algorithm and the realization of hardware functions,three groups of verification experiments were designed and completed,in which the detection accuracy experiment verified the accuracy of the clustering based target detection algorithm,the detection range experiment showed that the LiDAR could detect the range of the UAV,and the UAV dynamic flight experiment verified that the particle filter algorithm could track the UAV,and at the same time verified the successful transplantation of Linux operating system and the correct operation of the data acquisition program.更多还原