Non-metric distance judgements are influenced by image projection geometry and field of view

Abstract

Despite its mathematical simplicity and ubiquity in imaging technology, there has long been doubt about the ability of linear perspective to best represent human visual space, especially at wide-angle fields of view under natural viewing conditions. We investigated whether changes to image geometry had an impact on participants’ performance, specifically in terms of non-metric distance estimates. Our multidisciplinary research team developed a new open-source image database to study distance perception in images by systematically manipulating target distance, field of view and image projection using non-linear natural projections. The database consists of 12 outdoor scenes of a virtual 3D urban environment in which a target ball is presented at increasing distance, visualised using both linear perspective and natural perspective images, rendered respectively with three different fields-of-view: 100, 120 and 140 degrees horizontally. In the first experiment (N=52) we tested the effects of linear vs. natural perspective on non-metric distance judgements. In the second experiment (N=195) we investigated the influence of contextual and previous familiarity with linear perspective, and individual differences in spatial skills on distance estimations. The results of both experiments showed that distance estimation accuracy improved in natural compared to linear perspective images, particularly at wide-angle fields-of-view. Moreover, undertaking a training session with only natural perspective images led to more accurate distance judgments overall. We argue that the efficacy of natural perspective may stem from its resemblance to the way objects appear under natural viewing conditions, and that this can provide insights into the phenomenological structure of visual space.