【无人机】【2015.12】基于无人机视觉系统的飞机检测与跟踪

本文为美国安柏瑞德航空大学（作者：Yan Zhang）的硕士论文，共79页。

为了使无人机（UAV）能够在没有自动协作监视广播（ADS-B）的通用航空（GA）飞机等非合作飞行目标的国家空域安全飞行，无人机“看到”这些目标的能力尤为重要，或者可以通过“激光雷达”等多种方式实现。这里我们考虑只在无人机上安装摄像头，它具有重量轻、功耗低的优点。为了使视觉系统正常工作，要求基于摄像机的感知能力达到或超过人类飞行员的水平。

本文研究了基于摄像机的飞行器传感的两个基本问题/挑战。

第一个问题是，由于摄像头安装位置不同，无人机上拍摄到视频的稳定性。在这篇论文中，我们考虑了几种稳定化的演算法，包括卡尔曼滤波和粒子滤波。我们提供了这些滤波方法及其实现的详细讨论。

二是利用图像处理算法对飞机进行可靠的检测和跟踪。我们将形态处理和动态规划相结合，在不同的情况下取得了很好的效果。利用合成数据和记录数据对不同的图像处理算法进行了性能评估。

For unmanned aerial vehicles (UAVs) to operate safely in the national airspace where non-collaborating flying objects, such as general aviation (GA) aircraft without automatic dependent surveillance-broadcast (ADS-B), exist, the UAVs’ capability of “seeing” these objects is especially important. This “seeing”, or sensing, can be implemented via various means, such as Radar or Lidar. Here we consider using cameras mounted on UAVs only, which has the advantage of light weight and low power. For the visual system to work well, it is required that the camera-based sensing capability should be at the level equal to or exceeding that of human pilots.

This thesis deals with two basic issues/challenges of the camera-based sensing of flying objects. The first one is the stabilization of the shaky videos taken on the UAVs due to vibrations at different locations where the cameras are mounted.

In the thesis, we consider several algorithms, including Kalman filters and particle filters, for stabilization. We provide detailed theoretical discussions of these filters as well as their implementations. The second one is reliable detection and tracking of aircraft using image processing algorithms. We combine morphological processing and dynamic programming to accomplish good results under different situations. The performance evaluation of different image processing algorithms is accomplished using synthetic and recorded data.

1.   引言
   

2. 相机校正
3. 视频稳定性
4. 用于视频稳定性的粒子滤波和卡尔曼滤波
5. 物体检测算法
6. 结论

更多精彩文章请关注公众号：