Personal identification through facial recognition and visualizing details on HUD (P-0341) (MFN 8558)

Abstract

Facial Recognition is an effective technology and is used for biometric authentication and recognition of a person. It has a variety of applications. One of them is security. In this project smart glasses have been introduced that can perform facial recognition to serve as an aid in security measures. This smart device can easily capture the face frontal view which is difficult for security cameras and it is also portable. The techniques used to achieve face recognition are deep learning based due to their high accuracy as compared with old techniques like Eigen Faces, Principle Component Analysis, Local Binary Pattern Histogram, Fisher Faces and Linear Discriminate Analysis. Deep learning is subfield of machine learning that is training multiple layer neural network to perform a specified task. This device can easily recognize the person and is very effective in security and confidential areas for ensuring their safety. Facial Recognition is the matching of input image with a stored database and showing the result based on that comparison. Trained neural networks serve as a database for recognition. They compare the input image with trained model and predict the result. Human face pattern is an example of complex image. For dealing complex images Convolution Neural Networks (CNN's) are used. Transfer learning of CNN's can be done to achieve this task. In this project transfer learning of a pre-trained CNN "Face Net" is performed to achieve face recognition. Face Net has the highest accuracy among all CNN's while performing face recognition. Training of this network also requires a smaller number of images as compare to other convolution neural networks. The device hardware includes a camera mounted on glasses. It takes input in the form of image/video and send it to the processing unit (Raspberry Pi 3 B). The processing unit placed in the user's pocket perform the face recognition and show the output result on display (0.96 inches 0-LED) mounted on glasses, giving its projection on a reflector. Both camera and display are connected to processing unit wirelessly. The main objective is to enhance and improve the security system through this portable device.