基于边缘计算的紧致化掌纹识别系统

Design of a compact palmprint recognition system based on edge computing

随着生物特征身份认证技术的发展,无接触式掌纹识别的高识别率、低侵犯性和无接触性等优点,使其得到了越来越广泛的关注.为了满足边缘环境下多设备实时高效的处理需求,本文设计了一套基于边缘计算的紧致化掌纹识别框架,分别在终端设备层、边缘服务器层、云层搭建掌纹识别子系统.在终端设备层采用基于Tiny YOLO-v3的目标识别算法和基于MobilenetV2的关键点定位算法对采集图像进行预处理,提取掌纹ROI(region of interest),并提出识别请求.在边缘服务器层,对接受到的掌纹ROI利用基于对抗度量学习的GoogLeNet模型进行特征提取和特征匹配,在返回识别结果后进行数据同步.在云层,数据中心将会对所有的识别任务记录日志并归入数据库,同时定时训练更新终端设备和边缘设备的网络模型,以提高系统的跨领域识别能力.该框架是一套完整可行的生物特征识别框架,具有广阔的市场前景和应用价值.

With the development of biometric authentication technology,contactless palmprint recognition has gained increasing attention due to its high recognition rate,low invasiveness,and contact lessness.To meet the demand of real-time and efficient processing of multiple devices in the edge environment,this paper designs a compact palmprint recognition framework based on edge computing and builds a palmprint recognition subsystem in the terminal device,edge server,and cloud layers.In the terminal device layer,the Tiny YOLOv3-based target recognition algorithm and the MobileNetV2-based keypoint localization algorithm are used to pre-process the captured images,extract the palmprint region of interest(ROI),and make recognition requests.At the edge server layer,the received palmprint ROIs are extracted and matched with features using the GoogLeNet model based on adversarial metric learning.The data are then synchronized after the recognition results are returned.In the data center,all recognition tasks are logged and filed in the database.Moreover,the network models of the end and edge devices are regularly trained and updated to improve the system’s crossdomain recognition capability.The constructed framework is a complete and feasible biometric recognition framework with broad market prospects and application value.