Distance Estimation from a Monocular Camera Using Face and Body Features

The distance estimation of objects from a single camera can play a crucial role in many real-life applications, such as mobile robot navigation, autonomous vehicles and camera-based assistive devices for visually impaired people. This paper addresses the problem of distance estimation of people from a monocular camera. The paper proposes using face features, in addition to body features, with various regression algorithms to estimate the distance from a camera. These features are extracted from the bounding boxes of the people, detected in camera images. Two datasets, an ideal manual one and a practical automatic one, have been built for training and validation purposes. Various metrics, including a novel one proposed in this paper, have been utilized to evaluate the different regression models. The results reveal that the distance estimation errors have been significantly reduced using the body and face features together for all used regression algorithms, and a distance precision of about 98% has been achieved.