摘要
随着技术发展,智能门禁系统已经走入人们的日常生活中。更加智能、安全、低成本的门禁系统是智能家居的研究热点之一。以智能门锁为研究对象,以价格低廉的树莓派4B作为智能门锁核心控制单元,结合摄像头、智能开关锁、显示器等作为智能门锁硬件。同时,在识别算法中引入MTCNN+FaceNet,针对FaceNet与训练模型存在对中国人脸识别准确率不高的问题,对该模型在适用于中国人脸的数据集上,对该方法进行迁移学习再次训练。在LFW数据集上对模型进行评估,精度可达到98.6%。测试结果表明,该系统可以有效识别人脸并进行开锁操作,同时在Android APP中也可以进行远程操作,而且系统功能完整,识别准确率较高,价格低廉,在智能家居中有较高的使用价值。
With the development of technology,intelligent access control system has entered People’s Daily life.More intelligent,safe and low-cost access control system is one of the research hotspots of smart home.The intelligent door lock is taken as the research object,and the low-cost Raspberry Pi(4B)is taken as the core control unit of the intelligent door lock,and the smart door lock hardware is combined with the camera,smart switch lock and display.The recognition method of MTCNN+FaceNet is introduced into the recognition algorithm.Aiming at the problem that Facenet and the training model have low accuracy for Chinese people,the transfer learning is carried out to retrain the method on the data set suitable for Chinese faces.The model was evaluated on LFW data set,and the accuracy reached 98.6%.The test of the system shows that the system can recognize faces and unlock locks effectively.At the same time,it can also be operated remotely in the Android APP.The test results show that the system has the advantages of complete function,high recognition accuracy and low price.It has high application value in smart home.
作者
王博
WANG Bo(School of Electronic Information and Electrical Engineering,Shangluo University,Shangluo 726000,China;Artificial Intelligence Research Center of Shangluo,Shangluo 726000,China)
出处
《系统仿真技术》
2021年第1期34-36,42,共4页
System Simulation Technology
基金
商洛学院科研项目(19hxy155,16sky002)
