人体再识别系统的设计与实现  

Design and Realization of System of Person Re-identification

作　　者：刘钢[1] 曾明勇[2] 田畅[3] 吴泽民[3] 付毅[4] 

机构地区：[1]解放军理工大学通信工程学院研究生1队 [2]解放军理工大学通信工程学院研究生4队 [3]解放军理工大学通信工程学院 [4]中国人民解放军73667部队

出　　处：《军事通信技术》2017年第1期54-59,共6页Journal of Military Communications Technology

基　　金：国家自然科学基金资助项目(61501509)

摘　　要：人体再识别广泛应用于视频图像监控、公安图库检索和军事情报侦察等重要领域。文章设计实现了一种人体再识别系统,拍摄被关注人物的图像,即能在图库中查询得到相同目标的图像信息。系统由服务器和手机客户端组成,两者通过局域网组网。服务器建立关注人物的图像库,使用多通道级联特征(MCCF)算法实现目标的再识别;手机客户端采用改进的方向梯度直方图(HOG)算法,自动检测和提取拍摄场景中的行人,实现关注人物的入库和检索。经测试实验,客户端的行人检测能达到实时要求,系统再识别性能良好。Person re-identification is widely applied in video surveillance, image search in the police department, military information reconnaissance and other fields. A system of person re-i- dentification was designed and realized, and the target＇s identity was presented on the smart- phone in a short time once the target images were imported and required for retrieve. The system consists of one server and some smartphones as clients, which is connected with the LAN net-work using TCP/IP protocols. The server builds the database for the focused targets from local images and handles the requests of joining database and querying from the clients, and the server utilizes the multi-channel cascading feature algorithm with C＋ ＋. The client accomplishes its function of real-time pedestrian detection with improved HOG algorithm. A comprehensive per- formance evaluation was conducted which proves the good performance of the system.

关 键 词：再识别 行人检测 视频分析 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]