一种组网人脸识别门禁系统的设计

A Design of Access control system based on face recognition and Cloud technology

摘要:

针对传统人脸识别门禁中识别率不高的情况,设计了一种具有环境针对性的可靠人脸识别网络,采用卷积神经网络技术和基于TCP/IP的云应用技术,对人脸识别过程中的人脸提取进行了改进,采用了先进的MTCNN模型,实验结果表明,在2代和10代训练后识别结果收敛效果不明显,存在较大梯度,20代时训练后结果收敛,准确率逼近100%,loss值也逼近0%,表明20代训练模型确实已经完全收敛,且训练的拟合速度适中,既没有出现过快或者梯度消失的现象,也没有出现过慢或者不收敛甚至反升的情况,可以说模型的学习率也已经取到最佳效果。经工程测试,该系统具有更高的识别准确性和注册兼容性,可以满足小区、公司等人员数量不多且较为固定的场所的应用需求。

Abstract:

An environment adaptive face detection and recognition system is proposed in this paper to improve the recognition precision in traditional door access control systems. The two-stage convolutional neural network technology with precise division of labor and the cloud application technology based on TCP/IP is adopted to improve the performance of traditional systems. Experiment results, with the advanced MTCNN model, show that after 2 generations and 10 generations of training, it can’t reach the convergence, while the results are better after 20 generations of training. Practical tests demonstrate the potential applications in communities and companies where the number of occupants maintains is comparatively small due to the high precision and compatibility.