采用时空条件信息的动态场景运动目标检测  

Moving Object Detection in Dynamic Scene Using Spatial-Temporal Condition Information

作　　者：王斌[1] 肖文华[1] 张茂军[1] 熊志辉[1] 刘煜[1] 

机构地区：[1]国防科学技术大学信息系统与管理学院,长沙410073

出　　处：《计算机辅助设计与图形学学报》2012年第12期1576-1584,共9页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(60803101;60872150;61175006;61175015)

摘　　要：为了将动态场景中运动目标与扰动背景线性不可分的问题转换为线性可分问题,提出了负对数非线性核变换方法.该方法通过引入视觉注意机制构建视觉显著性时空域模型,以像素邻域加权条件信息作为分类特征,增强目标与背景的线性可分性,提高动态场景运动目标检测精度.最后结合图像分块建模策略,实现了动态场景中运动目标的高效、实时检测.This paper proposes a novel method to solve the non-linear classification problem of moving object detection in dynamic scene with spatial-temporal condition information. Our method converts the non-linear classification problem to a linear one by applying negative logarithm kernel transform algorithm. In order to get higher detection accuracy, a visual salient spatial-temporal domain is constructed to extract spatial-temporal condition information. Neighborhood weighted spatial-temporal condition information is employed for pixel classification. Combing with image block difference predetection, our proposed method can effectively detect moving object in dynamic scene in real time.

关 键 词：动态场景 运动目标检测 核变换 时空条件信息 视觉显著性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]