Face-to-camera distance estimation using machine learning

Abstract

Distance estimation of moving objects from the camera with accuracy is a challenging task in the digital era where human interaction increases with smart systems and state-of-the-art applications. The wide-ranging applications of distance estimation include the “zooming” effect in a document reader and finding the power of the correction lens for eyesight (eyesight testing). A lot of research has already been done on different methods of distance estimation like the most widely-used methods of "mono-vision" and "stereo-vision". The purpose of this study is to introduce a novel approach to finding the distance between the face and the camera with a high degree of accuracy and speed. The proposed method is based on detecting, measuring, and calculating the size of irises on an image of a human face obtained from a single camera, and a Supervised Machine Learning algorithm. The open-source Mediapipe Python package has been employed to extract the irises from the images. The proposed method has given the results of the estimated distance with an average accuracy of 95.6%.