Abstract: This paper introduces a novel method of detecting the moving maritme objects in video sequences. Those regions of interest are quickly segmented first and then a modified temporal differencing method is applied to detecting the moving objects. On the process of image segmentation, visual attention architecture was imported, the maritime noises are filtered successfully by building pyramid structure and making low-pass Guassian filter iteratively. Then those regions of interest are segmented successfully and quickly via boundary detection and analysis. Many experimental results show that the proposed real-time system produces satisfying results in various maritime video sequences.

Key words: Pyramid structure, Noise removal, Image segmentation, Moving objects detection, Temporal differencing

摘要： 本文提出了一种基于可见光成像序列的海上运动目标的快速检测方法。该方法先在静态图像中分割出感兴趣区域，继而仅在感兴趣区域中应用变形的时间差分法来检测运动目标。在感兴趣区域分割阶段，引入视觉注意力机制，通过构造金字塔结构以及迭代的小尺寸高斯低通模板运算成功地滤除了海面波浪噪声。在此基础上，检测并分析外边缘，在单帧图像中快速实现了感兴趣区域的分割。大量实验表明，该方法具有实时性好、鲁棒性强等特点。

关键词: 金字塔结构, 滤噪, 图像分割, 运动目标检测, 时间差分