Statistical body height estimation from a single image

We address the problem of estimating a person's body height from a single uncalibrated image. The novelty of our work lies in that we handle two difficult cases not previously addressed in the literature: (i) the image contains no reference length in the background scene to indicate absolute scale, (ii) the image contains the upper body part only. In a nutshell, our method combines well-known ideas from projective geometry and single-view metrology with prior probabilistic/statistical knowledge of human anthropometry, in a Bayesian-like framework. The method is demonstrated with synthetic (randomly generated) data as well as a dataset of 96 frontal images.

1. Introduction

In this paper, we re-visit the problem of body height (or stature) measurement from a single uncalibrated image. Body height estimation of people in images and video has many important applications, as body height can be used to identify individuals, either uniquely or partially. Applications where the camera calibration parameters are unavailable include forensics and detection/tracking of people from a moving camera or across multiple camera views. In forensic image analysis, body height can be used to rule out the possibility that a particular person is the same person in the image (i.e. screening and elimination of suspects) [4], [6], [18], [7]. In human detection/tracking applications, such as video surveillance and customer tracking, body height can be used to distinguish among a small set of tracked people in the scene [1], [16], [3], [15], [10].