为了在摄像机平台不稳的情况下获取稳定的图像序列,实现基于电子稳像的目标跟踪,对摄像机主动扫描运动与随机抖动分离方法进行了研究,提出了用均值偏移和粒子滤波结合的运动滤波算法(MSPF)来实现运动分离。算法通过粒子滤波预测粒子,然后利用单次均值偏移迭代移动粒子,使粒子更接近于目标真实位置区域,削弱了计算结果精度对粒子数的依赖。对于现实中的复杂背景,利用MSPF算法分离摄像机的主动扫描运动和随机抖动,采用运动补偿图像差分法检测出运动目标,从而实现图像的稳定跟踪。实验结果表明,MSPF算法使用50%的粒子就能起到传统粒子滤波算法同样的效果,缩短了计算时间,有利于实现实时稳像跟踪,适用于车载、船载、机载等稳定跟踪系统中。

In order to obtain steady image sequence from unsteady camera and to realize object tracking based on image stabilization,a separating method of active scanning movement from random jitter of camera are presented. Mean Shift with Particle Filter (MSPF) algorithm is proposed to realize motion separation. This algorithm uses particle filter to forecast particles, and then move them to be close to the real position of object through single mean shift iteration,which can undermine the dependence of the calculation accuracy on the number of particles. To a complex background in reality,the MSPF algorithm can be used to separate active scanning movement from random jitter of camera, and then image difference and motion compensation can be used to detect moving object to realize the stability of image tracking. Experimental results show that the MSPF algorithm with 50% particles can achieve the same effect with traditional particle filter,which means proposed algorithm can shortens process time and can realize real-time image stabilization for vehicle-borne, ship-borne, air-borne and other stable tracking systems.