航空图片中车辆的检测方法研究

本文提出了一种航空图片中车辆目标的检测方法。我们的检测过程分为三步，首先对图像进行旋转，使感兴趣的方向在图中为垂直方向，然后分别水平边缘和垂直边缘，利用车辆在航空图片中所呈现出来矩形轮廓特征对潜在的目标进行初步定位，然后提取的目标的区域投影特征对目标进行进一步的识别。算法利用了图像中目标的边缘特征和区域信息，可以进行完全自动检测识别，速度也较快，实验证明我们的算法对于航空图片中车辆的检测是很有效的。

Vehicle Detection in Aerial Image

An algorithm of detecting passenger cars in aerial image where cars appear as small objects is proposed. Our detection algorithm is based on the features of edge and area of a car. At first, directions of interest (DOI) of the image concerned are detected, the image is re-projected to a new plane to make the DOI vertical in image. Cars are primarily located by detecting right angles composed of vertical edge and horizontal edge. The projected features of the car are extracted to distinguish whether the object is a car or not. The whole process can run automatically and fast. The result is rather satisfactory when considering the low resolution of the aerial images.