引入区域种子的多运动目标分割算法  

Multiple Moving Objects Segmentation Algorithm Based on Regional Seed

作　　者：张琨[1,2] 王翠荣[1,2] 万聪[1,2] 刘冰玉[1,2] 

机构地区：[1]东北大学信息科学与工程学院,沈阳110819 [2]国家"九八五"工程下一代网络技术实验室,河北秦皇岛066004

出　　处：《小型微型计算机系统》2013年第8期1896-1899,共4页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(71071028)资助

摘　　要：针对智能监控系统中对多个运动目标进行图像分割这一问题,提出一种引入区域种子的多运动目标分割算法.算法首先利用背景减算法获得包含多个运动目标的前景图像,再利用四叉树分解方法获得与前景图像对应的稀疏矩阵,通过稀疏矩阵中数值的分布情况,计算出包含运动目标的区域种子点,从这些种子点出发,利用主动轮廓模型进行并行目标轮廓提取,最终完成多运动目标图像分割.实验结果证明本文算法能有效分割出前景图像中多个运动目标,分割结果与人眼视觉的判断相近,并行轮廓提取使算法具有良好的实时性.A multi-object segmentation algorithm with regional seed is proposed for the problem of multiple moving targets segmenta- tion in intelligent monitoring system. Firstly, it uses the background subtraction algorithm for foreground image containing multiple moving targets, and then uses quad-tree decomposition method to contain the sparse matrix corresponding to the foreground image, through the data distribution density of the sparse matrix, calculates the seed points corresponding to the regions which are containing the moving target, starting from these seed points, finally, execute the active contour model in parallel to complete the multiple mov- ing objects segmentation. Experimental results show that the proposed algorithm can effectively segment multiple moving targets and the segmentation results are close to the judgment of the human visual, parallel contour extraction makes our algorithm has good real- time.

关 键 词：四叉树 稀疏矩阵 区域种子 主动轮廓模型 多运动目标分割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]