FACE DETECTION AND RECOGNITION IN LOW RESOURCE ANDROID MOBILE DEVICES

Abstract

Nowadays, face detection and recognition are used in many mobile apps like attendance systems, access control, and login systems. But when we try to use these things in low-resource Android mobile phones, it becomes a bit difficult because mobile phones have less RAM, CPU, and battery, so we have to manage all these things properly. So, in my MTech thesis, I tried to make face detection and recognition work in such low-resource Android phones. In this project, for detecting faces, I used the Google ML Kit library. It is a ready-made Android library, which works fast and is already optimized for mobile phones. It can find faces in real time, draw boundaries on them, and crop the faces also. For recognizing the faces, I used two models — FaceNet and Mobile FaceNet. I converted both of these into TFLite format so that they can run easily on Android devices. I created a full Android app from scratch In the app itself when person register there face by using face detection and face recognition model his face embeddings is save and when he recognise it his embeddings are compared with the registered faces and whose embeddings are less is comes as detected face.