航拍图像中运动车辆的检测与定位

提出了一种利用Huber函数和均匀稀疏采样估计仿射变换参数的方法，更加准确快速地进行背景运动补偿,为了克服噪声和和伪运动区域对运动目标定位的干扰，利用形态学滤波方法去除法照变化带来的伪运动区域及噪声。最后，采用仿射运动矢量聚类方法对运动目标进一步可靠定位，结合这3种方法，可以对航拍图像序列中的运动目标快速准确地检测与定位。

Moving Vehicles Detection and Location in Aerial Image Sequence

A new method is proposed to estimate the affine transform model parameters, which is based on sparsely and evenly sampling to accelerate the calculation. Furthermore, Huber function is used to separate the background pixels from the moving vehicles' pixels according to a error measure. This method is sensitive to background pixels, so a more precise parameters' estimation can be got. To detect the true moving areas, an erosion filter is taken to remove the pseudo-moving areas and noise, and then a dilation filter merges the neighbors in order to get a complete moving area. A further robust approach to clustering moving areas belonging to the same movingobjects is based on affine motion vector clustering techniques. Combining the three methods, the moving vehicles in aerial image sequence can be detected and located precisely.