REAL TIME FACE DETECTION AND RECOGNITION FROM VIDEO USING DEEPFACE CONVOLUTIONAL NEURAL NETWORK

Abstract

Purpose: To implement secured and contactless in-store order pickup system based on real time face recognition to ensure the authenticity of the consumer and also to reduce the risk of frontline workers who are vulnerable to the prevailing COVID – 19.

Methodology: Online shopping website is developed using HTML/CSS. QR Code is generated for the corresponding order ID. In this system the face detection and recognition are done using Haar Cascade Classifier and Convolutional Neural Network Algorithm. 2D Convolution is used to train the model. 3 layers of convolution is used to get a testing accuracy of 98.99% and validation accuracy of 94.76%. ReLU and Softmax activation functions are used in this system. Structural Similarity Index is used to compare faces and get the desired output.

Findings: Input layer in CNN ought to contain picture information. Picture information is addressed by three-dimensional grid as we saw before. You need to reshape it into a solitary section. On the off chance that you have "p" preparing models measurement of information will be (625, p). Convolutional layer is now and again called include extractor layer since highlights of the picture are get separated inside this layer. Above all else, a piece of picture is associated with Convolutional layer to perform convolution activity as we saw before and ascertaining the dab item between open field (it is a nearby locale of the information picture that has the very size as that of channel) and the channel. Pooling layer is utilized to diminish the spatial volume of info picture after convolution. It is utilized between two convolution layers. The Euclidean distance or Euclidean measurement is the customary distance between two focuses that one would quantify with a ruler, and is given by the Pythagorean recipe.

Originality/value: The training accuracy acquired during training is 98.99 percent, and the validation accuracy is 94.76 percent. Thus, the CNN model may be utilized to detect and recognize faces accurately from any given video.