Abstract
Renaissance artists noticed that placing objects on a visible ground plane anchors them stably, making it easy to perceive their depth. Subsequently, they developed methods for geometric calculation of perspective by drawing construction lines defining the ground plane. Thus, the artist constructs the geometry of pictorial space, based on the station point and the view direction, then places objects in it. This practice created a scale for placing and sizing objects on a minimally patterned ground plane. Our experiments, which measure the precision of depth perception in perspective images, hypothesize that spatial perception evolved in the presence of two constancies, vertical gravity and an almost horizontal ground. Human sensitivity to horizontal and vertical orientations exemplifies the important effect that gravity and the ground have on human perception. From them a third constancy emerges, objects in contact with the ground at the same distance lie along a horizontal line, which artists' floor constructions highlight. In psychophysical experiments we measured the response time and accuracy of forced-choice closer/farther judgments between two objects placed in simple scenes based on traditional artist's perspective. The presence or absence of ground constructions and its orientation are varied in SVG images to minimize pixellation artifacts. Among the directions of gravity participants best perceive depth when the ground plane is horizontal and gravity downward, as measured by response time at 97% accuracy. Furthermore, the horizontal construction lines, which abstract the horizontal guides present in Renaissance paintings the tiled floors, improve depth perception substantially. In Renaissance scene paintings feet are usually visible; our experiments show that the feet provide the viewer with an accurate perception of relative depth, configuring the scene elements into clusters. The underlying cause lies in the structure of vision, which privileges the horizontal ground and downward gravity.
Meeting abstract presented at VSS 2018