一种基于统计模型的前景阴影消除算法  

Shadow Elimination Algorithm for Foreground Based on Statistical Model

作　　者：杨志邦[1,2] 徐成[1] 周旭[3] 田峥[1] 

机构地区：[1]湖南大学计算机与通信学院,长沙410082 [2]中国电子科技集团公司第三十六研究所通信系统信息控制技术国家级重点实验室,浙江嘉兴314033 [3]嘉兴学院数学与信息工程学院,浙江嘉兴314001

出　　处：《小型微型计算机系统》2013年第2期423-428,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60973030)资助;浙江省自然科学基金(LY12F02019)资助;浙江省公益技术应用研究项目(12011C23130)资助

摘　　要：视频图像中存在的阴影是影响运动目标检测效果的关键因素之一,对阴影进行检测和消除已成为运动检测中的重要研究内容.针对阴影消除问题,本文采用直方图统计方法,将阴影特征引入到传统混合高斯模型中,基于统计特征建立阴影高斯模型;在模型基础上,提出一种新的前景阴影消除算法,将前景像素与阴影模型进行匹配,实现阴影的判定和消除.与同类算法的对比分析表明:本文算法对于不同场景下的阴影消除是准确且实时的,在阴影检测率和阴影区分度上均有显著提升.The existence of the shadow is one of the key factors which impact the result of the object detection, detecting and elimina- ting the shadow become the important research area in the target detection. Aiming to the shadow problem during the object detection, we used the histogram to statistic and analysis the color feature of the shadow under the HSV color space, got the shadow feature of the H, S, V channels. Then established the Gaussian shadow model on each channel according to the statistic information. Based on the model had built, we proposed a novel algorithm to eliminate the shadow, using the foreground pixels to matche the model then de- termine and eliminate the shadow. Compared with the similar algorithms, the results show that the proposed algorithm can eliminate the shadow correctly in real-time under different scenarios, and it performs better on the metrics of shadow detection rate and the shad- ow discrimination rate.

关 键 词：运动目标检测 统计特征 高斯模型 前景 阴影消除 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]