多摄像头视频监控系统中基于SIFT特征的目标匹配研究  

An approach for object matching based on SIFT feature in multiple-camera surveillance system

作　　者：王津[1] 张学杰[1] 

机构地区：[1]云南大学信息学院,云南昆明650091

出　　处：《云南大学学报（自然科学版）》2009年第S1期37-42,共6页Journal of Yunnan University(Natural Sciences Edition)

基　　金：云南大学学术创新团队建设计划

摘　　要：多摄像头视频监控系统中,如何进行来自多个摄像头之间的数据融合是非常重要的问题.本文提出一种基于SIFT特征匹配进行多摄像头视野下的持续运动目标跟踪方法.根据SIFT特征对图像旋转、尺度缩放、亮度变化保持不变性,以及对视角变化、仿射变换、噪声也保持一定程度稳定性的特点,提出了基于SIFT的目标匹配算法.并通过对提取关键点、选取匹配阀值的改进,使算法达到了实时性要求.实验表明,通过图像之间目标的SIFT特征进行一致性匹配能有效地实现广域监控场景中多摄像头的协同跟踪,具有很好的跟踪性能.In a multiple-camera video surveillance system,how to integrate all data that come from several cameras is a very important problem.It is presented a sustained object track method based on SIFT feature matching.According to the characteristic that SIFT feature keeps invariant in the case of rotating picture,zooming scale,or changing brightness,and maintains stable under changing perspective,affine transformation or noise,we proposed an algorithm for object matching based on SIFT.In order to meet real-time requirements,we also improved the extraction of key-point and selection of match threshold.Experiment showed that using SIFT feature for consistence matching can enable cooperatively object track through multiple cameras in a wide areas video surveillance system,and had good track performance.

关 键 词：多摄像头系统 目标匹配 SIFT特征 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]