QQ客服 官方微博 反馈留言

	基于两级时空随机取样的背景建模算法
	熊祺 ; 张旭 ; 王生进 ; Xiong Qi ; Zhang Xu ; Wang Shengjin
	2016-03-30 ; 2016-03-30
关键词	时空随机采样 背景建模 前景检测 两级决策 temporal-spatial random sampling background modeling foreground detection two-leveldecision TP391.41
其他题名	Background Modeling Algorithm Based on Two-level Temporal-spatial Random Sampling
中文摘要	目的为从发射场室外和航天器舱内等场景的监控视频中提取运动目标,本文提出两级时空随机取样背景建模算法。方法提出了一种全新的背景建模算法,算法通过时间和空间上随机取样的方式进行背景更新,并且通过负反馈的方式实时调整参数。在前景决策上采用基于像素点和基于区域分布两级决策方式进行,充分利用了连续图像序列时间和空间上的一致性和随机性。结果在真实场景和Change Detection数据库上的测试表明本方法提取的前景稳定、清晰,召回率等指标上得到更好的结果。结论通过本文提出的方法,可直接从监控视频中提取较清晰的运动前景,在发射场和未来飞行器监控系统中将有较好的应用前景。; Objective To propose a two-level temporal-spatial random sampling background modeling method algorithm for extracting moving objects from the surveillance videos of spacecraft and launching site. Method A novel background modeling algorithm was presented. In our algorithm,the background was updated through random temporal-spatial sampling. Parameters were dynamically adjusted according to the negative feedback information. The foreground decision was performed in a two-level decision manner based on pixels and regional distribution,making full use of the temporal-spatial consistency and randomness of continuous image sequence. Results The approach was evaluated on the real surveillance video and the database provided for the Change Detection Challenge. Experiments showed that compared with other modeling methods,better performances could be achieved using the proposed algorithm. Conclusion With the background modeling algorithm, clear foreground directly extracted from the surveillance videos can be used for early-warning and action recognition from the surveillance videos of the spacecraft and launching site.
语种	中文 ; 中文
内容类型	期刊论文
源URL	[http://ir.lib.tsinghua.edu.cn/ir/item.do?handle=123456789/146656]
专题	清华大学
推荐引用方式 GB/T 7714	熊祺,张旭,王生进,等. 基于两级时空随机取样的背景建模算法[J],2016, 2016.
APA	熊祺,张旭,王生进,Xiong Qi,Zhang Xu,&Wang Shengjin.(2016).基于两级时空随机取样的背景建模算法..
MLA	熊祺,et al."基于两级时空随机取样的背景建模算法".(2016).

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。