【信息技术】【2004】基于计算机视觉的无人机自主避障系统中的目标跟踪研究
本文为瑞典皇家理工学院（作者：Johan Driessen）的硕士论文，共68页。
这篇硕士论文的目的是研究利用商用货架(COTS)硬件和免费、公开可用的计算机视觉库开发一个足够有效的实时自主避障系统。介绍一些用于自主避障系统的跟踪和数据关联的技术，一些算法已经得到实现并用于演示。论文中的一部分也致力于研究目前可用视频硬件的局限性。
本文主要针对无人机自主避障的跟踪问题进行了研究。在Staffan Rydergard的硕士论文《无人机自主防撞系统中的障碍物检测》对障碍检测问题进行了研究，他与本文作者在基础方面进行了合作。强烈建议将这两篇论文一起阅读，以便更全面地看待整个系统的问题。
本文仅考虑以下飞行情况：
-
无人机保持水平飞行（以恒定的姿态角和飞行速度）； -
无人机在高空飞行； -
只考虑无人机视线中出现在地平线上方的物体； -
来自无人机导航系统的精确数据可用于自主避障系统。
本文的主要成果是在个人计算机上实现了多种卡尔曼滤波器的跟踪算法。然而，标准摄像机的分辨率有限，很难成功地开发出性能足够好、能够被民用空域接受的自主避障系统。在可预计的不久的将来，具有更高分辨率的视频系统在市场上将变得更加常见，因此使用COTS硬件的基于计算机视觉的自主避障系统具有很大的研究潜力，应当鼓励在该领域开展更多的工作。
The purpose of this master’s thesis is toinvestigate the possibilities of developing a sufficiently effective real-timesee-and-avoid system using Commercial Off-The-Shelf (COTS) hardware and free,publicly available computer vision libraries. This has been done primarily inan exposition of some interesting techniques for tracking and data associationavailable for use in a see-and-avoid system. A limited implementation of some algorithmshas also been made for demonstration purposes. A part of the thesis has alsobeen devoted to examining the limitations that come from the video hardwareavailable. This thesis is focused mainly on the tracking aspect of thesee-and-avoid problem for unmanned aerial vehicles (UAVs). The detection aspectis examined in the master’s thesis by Staffan Rydergård, “Obstacle Detection ina See-and-Avoid System for Unmanned Aerial Vehicles”, who has been workingtogether with the author of this thesis on the basics. It is highly suggestedthat both theses are read together for a more complete view of the problem. Inthis thesis, only the following flight situation is considered: • The UAV willbe in level flight (with constant attitude angles and airspeed) • The UAV willfly at high altitude • Only objects that appear above the horizon from theUAV’s line of sight will be considered. • Accurate data from the navigationsystem of the UAV will be available to the see-and-avoid-system. The primaryresult of this thesis is that there are several tracking algorithms, primarilyversions of the Kalman filter, that should be adequate to be able to implementsuch a system on personal computers. However, the limited resolution ofstandard video cameras makes it very difficult to successfully develop asee-andavoid system that can perform well enough to be acceptable for civilianairspace using these. Since video systems with higher resolutions are expectedto become more common on the market within the near future, computervision-based see-and-avoid systems using COTS hardware nevertheless has thepotential of becoming a very interesting research field, and more work in thisarea is encouraged.
1 引言
2 自主避障系统基础
3 图像分析
4 目标跟踪
5 数据关联
6 目标分类
7 碰撞风险确认
8 算法实现
9 结论与讨论
10 参考文献
附录1 光学关系
附录2 避障时间关系
附录3 XC-003P相机技术规格
附录4 飞机性能参数
附录5 地平线确定
下载英文原文地址：
http://page2.dfpan.com/fs/9l4c6j423251b2b9163/
更多精彩文章请关注微信号：