基于内容的海量监控视频的多层次检索系统  

Content-based Multi-level Retrieval System for Massive Surveillance Video

作　　者：郑海波[1] 韩小萱[1] 史云静[1] 李洁[1] 朱秀昌[1] 

机构地区：[1]南京邮电大学江苏省图像处理与图像通信重点实验室,江苏南京210003

出　　处：《电视技术》2014年第19期196-201,共6页Video Engineering

基　　金：国家自然科学基金项目(61071091)

摘　　要：设计和实现了一种基于内容的海量监控视频的多层次检索系统。该系统首先从监控视频中提取关键帧图像,其次利用行人检测、人脸识别及车辆检测等算法将关键帧中的行人图像、人脸图像和车辆图像等感兴趣目标提取出来,然后提取这些图像的颜色、纹理等特征,利用改进的LIRe(Lucene Image Retrieval)建立分布式的特征库,最终形成了多层次的信息数据库。实验表明,该系统具有较高的检索准确率和较快的检索速率,并支持海量监控视频的检索。In this paper, a content-based multi-level retrieval system for massive surveillance video is design and realized. Firstly, key frames are selected from the surveillance videos. Secondly, interested targets, which include pedestrians, faces and cam, are segmented from the chosen key frames through human detection, face recognition and vehicle detection correspondingly. Finally, features like color or texture of these object images are utilized to construct a distributed feature library via improved LIRe （Lucene Image Retrieval）. In this way, a multi-level database is establishcd. Experiment results show that the proposed system performes well on both precision and efficiency, as well as supports the retrieval for massive surveil- lance video.

关 键 词：多层次 监控视频 LIRe 关键帧 

分 类 号：TN919.85[电子电信—通信与信息系统]