Abstract
The amount of multimedia content is growing exponentially and a major portion of multimedia content uses images and video. Researchers in the computer vision community are exploring the possible directions to enhance the system accuracy and reliability, and these are the main requirements for robot vision-based systems. Due to the change of facial expressions and the wearing of masks or sunglasses, many face recognition systems fail or the accuracy in recognizing the face decreases in these scenarios. In this work, we contribute a real time surveillance framework using Raspberry Pi and CNN (Convolutional Neural Network) for facial recognition. We have provided a labeled dataset to the system. First, the system is trained upon the labeled dataset to extract different features of the face and landmark face detection and then it compares the query image with the dataset on the basis of features and landmark face detection. Finally, it compares faces and votes between them and gives a result that is based on voting. The classification accuracy of the system based on the CNN model is compared with a mid-level feature extractor that is Histogram of Oriented Gradient (HOG) and the state-of-the-art face detection and recognition methods. Moreover, the accuracy in recognizing the faces in the cases of wearing a mask or sunglasses or in live videos is also evaluated. The highest accuracy achieved for the VMU, face recognition, and 14 celebrity datasets is 98%, 98.24%, 89.39%, and 95.71%, respectively. Experimental results on standard image benchmarks demonstrate the effectiveness of the proposed research in accurate face recognition compared to the state-of-the-art face detection and recognition methods.