分层级联结合跟踪反馈的复杂场景目标检测  

Hierarchical Cascade Combined Tracking Feedback for Target Detection in Complex Scenes

作　　者：曹倩霞[1,2] 罗大庸[1] 王正武[2] 

机构地区：[1]中南大学信息科学与工程学院,长沙410075 [2]公路工程省部共建教育部重点实验室(长沙理工大学),长沙410004

出　　处：《小型微型计算机系统》2014年第8期1901-1905,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(51278068)资助;湖南省科技计划项目(2012GK3060)资助;长沙理工大学公路工程省部共建教育部重点实验室开放基金项目(kfj100105)资助

摘　　要：针对复杂场景中非平稳背景扰动问题和暂停(或低速)运动目标容易污染背景问题,提出一种块级-像素级分层级联背景差分和跟踪反馈相结合的目标检测方法.首先,基于Sigma-delta背景算法,由能有效处理非平稳背景的块级背景差分快速检测粗目标,然后基于Camshift-Kalman跟踪算法,对粗目标进行跟踪并反馈状态信息至块级背景差分算法中约束块级背景更新和像素级前景提纯操作以阻止暂停(或低速)运动目标污染背景,同时,将块级背景更新状态信息反馈回到跟踪算法中约束跟踪操作以提高跟踪精确性,二者相互作用.实验表明,该方法在存在非平稳背景或暂停(或慢速)运动目标等复杂场景中能实现鲁棒的前景检测效果.To solve the problems of non-stationary background disturbance and temporarily stopped moving （ or slow-moving ） targets easily contaminated background in complex scenes, a target detection method combined block-level and pixel-level hierarchical cas- cade for background subtraction with tracking feedback was presented. First, based on Sigma-delta background algorithm, the coarse targets are quickly detected by the block-level background subtraction which can effectively deal with non-stationary background. Then, based on Camshift-Kalman tracking algorithm, the coarse targets are tracked and its status information is fed back to the block- level background subtraction algorithm to constrain the block-level background updating and pixel-level foreground refining operations to prevent temporarily stopped moving （or slow-moving ） targets contaminated background, while the status information of the block- level background updating is fed back to the tracking algorithm to constrain tracking operation to improve tracking accuracy, there are interactions between them. Experiments show that the method can achieve robust foreground detection results in complex scenes with non-stationary background or temporarily stopped moving （ or slow-moving ） targets.

关 键 词：目标检测 背景差分 Sigma-delta滤波 Camshift-Kalman跟踪 复杂场景 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]