用于海上感兴趣区域实时分割的近似算法

The Approximation Algorithms Applied to Maritime Region of Interest Segmentation

详细介绍了海上运动目标实时检测和跟踪系统中感兴趣区域分割阶段所使用的主要算法及其优化策略.主要分析了系统所涉及的两个方面的工作:(1)引入视觉注意力机制,在构建金字塔结构的基础上,提出了一种迭代的基于倒三角小模板的线性低通滤波方法,快速地实现粗分辨率图像上的平滑滤噪,凸显目标;(2)用近似区域填充算法或是水平放置的最小外接矩形逼近存在较大缺口的边缘连通分支,以构成物体的大致区域.理论分析表明,这些近似算法在提高运算速率的同时,仍然具有较高的近似比.同时实验结果也表明,算法具有实时性好,鲁棒性强等特点.

This paper introduces the approximation algorithms applied to the progress of region of interest segmentation of realtime moving maritime object detection and tracking system. Two parts of the system are analyzed mainly： 1. Visual attention architecture is imported to reduce the maritime noises and make the object distinct from the background by building pyramid structure and making low-pass inversed triangular filter of small mask iteratively; 2. The approximative region of the object is constructed by either filling the region approximatively or constructing the minimum external rectangle according to the embranchment of the connected edge. Those approximation algorithms not only enhance the speed but also guarantee the high approximation ratio. Many experimental results show that the algorithms produce satisfying results in various maritime video sequences.