FACE MASK DETECTION USING CNN AND FROM DATASET TO DEPLOYMENT : CHALLENGES IN IMPLEMENTATING FACE MASK DETECTION CNNs IN PUBLIC TRANSPORT SYSTEMS

Abstract

Following the pandemic, it has become clear how much better it is to wear a face mask in crowded and closed areas like bus and train stations or other public venues. Manually ensuring that people are wearing masks in these places is very difficult and costly. A Convolutional Neural Network (CNN) is used in this thesis for real-time detection of faces with masks on public transport. To start, the study collects and curates images that feature faces in various lighting, orientations, occluded or with masks and in different situations. An artificial neural network model is set up, given data, taught how to classify and assessed for the task of telling whether someone is wearing a mask on this dataset. Model effectiveness is based on accuracy, precision, recall and F1- score. Along with developing algorithms, this thesis also examines problems related to putting them into real use. Among the things it includes are poor computation on devices at the edge, changes in lighting inside vehicles, motion-induced blur in pictures from surveillance and ethical issues related to video surveillance. The study investigates lighter models, explores ways to improve preprocessing and examines how to connect the technology to CCTV currently in place. This work details the entire procedure needed to detect face masks in transit locations. It shows how making deep learning work with low resources and fast updates is possible without putting either performance or feasibility at risk. The thesis offers a system model and installation framework that addresses hardware, privacy and scalability, to help move laboratory findings into general use.