基于小波多尺度图像配准的运动小目标检测

Detecting small moving object based on wavelet multi-scale image registration

探讨了一种红外图像序列中独立运动小目标检测的新方法。通过一种鲁棒的小波多尺度图像配准过程消除主场或背景运动的影响。同时,利用改进的Fisher算法及小波变换对低频的图像进行分割,最终将目标从背景中分离出来,并对配准后分割的图像差分,获得目标运动轨迹。实验表明了该方法的有效性。

A novel system to independently detect small moving object in IR image sequences taken from a moving platform is proposed. Egomotion effects are removed through a waveletbased robust multiscale affine image registration process. The low frequency images are segmented with the improved Fisher algorithm and wavelet transformation, then the object is separated from background. The difference of the segmented image after registration is computed, and the kinematic track of the object is obtained. The experimental result shows that the system is effective.