摘要
传统的集中式人脸识别方法在时间效率和可扩展性等方面存在不足,已经不能满足大规模实时人脸识别的需求。针对这个技术瓶颈,本文提出了一种分布式人脸识别方法,该模型由多个代理和一个服务器组成。代理能够同时对多个视频中的行人进行检测、跟踪以及特征提取,服务器则对视频中的行人执行识别操作。针对代理处理的任务分布不均而导致处理视频时间过长、任务量过大引起的CPU利用率激增问题,设计了代理的负载均衡来进行性能优化。利用代理统计处理的视频总数及每个视频中的行人数,并将统计数据发送给服务器。服务器通过负载均衡将视频重新分配给每个代理进行人脸识别。实验结果证明,分布式人脸识别有效地提高了人脸识别方法的效率和可扩展性。对一些较为极端的实验例子,经过性能优化后,代理中最大的CPU利用率降低近40%,有效地缓解了时间延迟问题。
Since the scale of video to be monitored and increasing, the traditional centralized face recognition processed by face recognition system has been methods are insufficient in time efficiency and scalability and no longer able to meet the needs of large scale and real-time face recognition. Aimed at such technical bottleneck, a distributed face recognition method was proposed. The model consisted of several agents and one server, in which the agent was able to detect, trace and extract features of pedestrians in several videos at the same time, while the server was able to identify those pedestrians in videos. Subject to the problems caused by uneven distribution of agent processing tasks, including long processing time, too much tasks and CPU usage explosion, a load balance of agent was designed for performance optimization. Firstly, the agent was used to count the total videos to be processed and number of pedestrians in each video, then, statistical data was sent to the server through the agent and at last, the server will re-allocate the videos to each agent for face recognition by the load balance.The results show that the distributed face recognition method can effectively promote the efficiency and scalability of face recognition methods. For those extreme cases, after performance optimization, the maximum CPU using rate in agent has been declined by approximate 40 %, which could effectively alleviate the time delay.
作者
闵卫东
石杰
韩清
王玮
MIN Wei-dong SHI Jie HAN Qing WANG Wei(School of Computer Science and Software Engineering, Tianjin Polytechnic University, Tianjin 300387, China School of Information Engineering, Nanchang University, Nanchang 330031, China)
出处
《光学精密工程》
EI
CAS
CSCD
北大核心
2017年第3期779-785,共7页
Optics and Precision Engineering
基金
国家自然科学基金资助项目(No.61302127)
江西省自然科学基金重大项目(No.20161ACB20004)
关键词
人脸识别
分布式
代理
负载均衡
性能优化
face recognition
distributed
agent
load balance
performance optimization
