为了实现对卫星视频图像序列的运动目标进行跟踪，提出了一种背景建模与添加实际条件约束相结合的卫星视频运动目标检测算法。首先，针对卫星视频图像序列的背景建立背景模型。然后，把接下来各帧图像同该模型相比较，进行减法运算，所得偏差为运动区域。最后添加满足卫星实际情况的约束条件，根据卫星视频的分辨率估算出运动目标合理的像元尺寸,从而滤掉噪声。将本算法在PC端进行实现，并与三帧差法和背景建模法的检测准确度进行对比。实验结果表明：本算法对卫星视频图像序列运动目标检测的准确度可达83.6%，三帧差法检测准确度为6.1%，背景建模法检测准确度为21.0%。从实验结果可以看出，本文算法可以比较准确地对卫星视频图像序列进行运动目标检测。

In order to realize the moving objects detection on the satellite video, an algorithm based on background subtraction method and actual physical constraints are proposed. First, we establish a background model representation of the scene. Then each frame compares with the model and makes a subtraction operation. The result is the motion region. Finally, we use actual physical constraints to reduce the noise because moving objects’ pixels can be estimated by satellite’s resolution ratio. We accomplish our method in PC, and compare their accuracy. Experimental results indicate that our method’s accuracy is 83.6%, three-frame difference method’s accuracy is 6.1%, background subtraction method’s accuracy is 21.0%. It can satisfy our method can accurately detects moving objects on satellite video.