基于时空联合的水下运动目标检测方法

提出了一种基于时空联合的运动目标检测算法,考虑到四阶累计量对于噪声具有较好的抑制作用,利用帧差的四阶累计量来提取运动区域(运动模板),与传统基于背景噪声方差的阈值确定方法不同,采用单帧图像上的有效带宽作为阈值,从而消除诸如水草、海洋生物等引起的背景纹理缓慢变化.为了得到准确的分割对象,引入马尔可夫随机场,结合区域的时空属性,把对象分割问题转化为求取区域最小能量问题,对水下噪声起到了较好的抑制作用.实验结果表明该算法能够有效地检测出具有精确边缘的水下运动目标.

Underwater object detection based on spatio-temporal combined algorithm

This paper proposed a moving object detection method based on spatio-temporal information.As the fourth-order cumulant had a good noise rejection capability,it was therefore possible to extract the moving region(motion mask) by the use of four-order higher order statistics.Compared with traditional threshold method based on background variance,calculated was the effective bandwidth as the threshold in single frame,which can eliminate the background texture change.To be get precise target segmentation,Markov random field which combining the spatio-temporal attributes and translated segmentation problem into region energy minimum problem,which played a better inhibition of underwater noise was introduced.The experimental results show that the proposed algorithm can effectively detect the moving target.