Abstract
The COVID-19 pandemic has proven the presence of preventative measures such as social distancing to be essential to public life. Even with options for vaccination, the recommendation for a 6-foot guideline still carries much relevance -especially in public, crowded spaces. Thus, the aim of this study is to ultimately develop a social-distancing “detector” through the comparison of two models: a model implementing the Gaussian thin-lens formula and a machine-learning model. Both models were tested for accuracy in detecting distance from a first-person webcam. Three hundred images were taken of a human subject at different distances, 58 of which were used for testing the models and 242 to train the machine-learning model. Results show that the testing accuracy of the Gaussian Thin-lens Model was 0.3966, or 39.66%. The machine-learning model achieved a higher testing accuracy of 0.9483, or 94.83%. This is likely since the Gaussian Thin-lens Model requires strict conditions to be met in order to calculate distances, one being the ability to locate facial landmarks to obtain a reference point. In contrast, the machine-learning model is able to adapt, making it the optimal choice for detecting distance from a webcam. It is possible that the model could be used to enforce or detect social distancing.