一种基于级联卷积网络的移动端人脸识别系统实现

A Face Recognition System on Mobile Devices Based on Cascaded Convolutional Networks

考虑到现有人脸识别算法的准确率和可移植性,设计并且实现了一个基于Android平台的人脸识别考勤系统。该系统可以直接部署到采用Android系统的移动设备,通过摄像头采集图片并根据已录入的人脸库数据对检测到的人脸进行识别和考勤登记。本系统采用一种基于级联卷积网络的人脸检测算法MTCNN和人脸特征提取算法FaceNet来实现人脸识别。考虑到移动设备计算能力的有限性,采用NCNN框架将两种算法进行移植到Android平台。最终测试表明,本文系统的实时性和准确率基本满足实际需要。

In this paper,according to the accuracy and portability of existing face recognition algorithm,a face recognition system which based on the Android platform is used in attendance.The system can take photos and identify the registered people who in the database,and it can be directly deployed to mobile devices using the Android system.In the algorithm,the face detection algorithm called MTCNN based on cascaded convolutional network and Face feature extraction algorithm called FaceNet are used to realize the face recognition.Considering the limited computing power of mobile devices,the deep learning framework called NCNN was used to transplant algorithms to the Android platform.Finally,the accuracy and real time of the system in this paper meet the actual needs.