Height measurements in images: how to deal with measurement uncertainty correlated to actual height

Abstract

The accuracy of body height estimations in images may be validated by performing height measurements on test persons of known height on images taken under similar conditions. The observed differences between actual and measured heights of test persons are then used to adjust the measurement on the perpetrator, leading to a confidence interval for his actual height. The statistical model used in casework assumes that the actual heights of test persons and the differences of measured and actual heights are uncorrelated. In an experimental set-up, it is investigated whether there is a linear dependence between actual heights and differences between actual and measured heights. The assumption of independence of differences to actual heights is violated in one out of four experimental cases. For cases in which a linear dependence is present, an alternative statistical model is described for determining confidence intervals. A step-by-step illustration is given of implementation of the data analysis proposed both for a linear and for a restricted linear model.