Abstract: A detection algorithm of moving objects under moving camera is presented.Firstly，select feature points from two adjacent pictures and put their coordinate in affine transform mode to get global motion vector discribing the moving background.Secondly，use this affine model to match the two adjacent frame backgrounds.Finally，perform a subtraction between these two matched adjacent frame backgrounds for detection of the moving object.The algorithm is not only superior to original algorithms in velocity and accuracy but also can deal with larger rotational moving of the camera.

摘要： 提出一种摄像机运动下的运动目标检测技术。首先获取相邻两帧图像上的对应特征点的坐标后应用到仿射变换模型得到全局运动参量来描述背景图像的运动变化。然后根据估计出的全局运动参量通过双线性内插法完成背景匹配。最后采用背景匹配后的两帧图像差进行目标检测。实验证明本算法在提高处理的精确度和时效性的同时还可以处理摄像机发生较大幅度旋转运动情况下的目标检测。