Evaluation of depth perception in crowded volumes

TL;DR

This work investigates depth perception in crowded volumes where occlusion is prevalent, emphasizing camera motion as a key perceptual cue beyond rendering technique. An online crowdsourced study compares four rendering methods (two traditional: EAM, ISO; two physically based: DOS, VPT) across four volumes with varying crowdedness, using a circular camera-motion task and mean absolute depth error $MAD$ as the metric. Results show depth perception is driven more by content and motion cues than by rendering technique, with $MAD$ showing no significant method differences but strong volume effects; crowdedness degrades performance, particularly for very dense small-structure volumes, while larger instances and motion parallax improve accuracy. The findings highlight the importance of interaction and content-aware sparsification for crowded-volume visualization and point to future work on enabling direct user control over sparsification and grouping to sustain depth perception in complex data.

Abstract

Depth perception in volumetric visualization plays a crucial role in the understanding and interpretation of volumetric data. Numerous visualization techniques, many of which rely on physically based optical effects, promise to improve depth perception but often do so without considering camera movement or the content of the volume. As a result, the findings from previous studies may not be directly applicable to crowded volumes, where a large number of contained structures disrupts spatial perception. Crowded volumes therefore require special analysis and visualization tools with sparsification capabilities. Interactivity is an integral part of visualizing and exploring crowded spaces, but has received little attention in previous studies. To address this gap, we conducted a study to assess the impact of different rendering techniques on depth perception in crowded volumes, with a particular focus on the effects of camera movement. The results show that depth perception considering camera motion depends much more on the content of the volume than on the chosen visualization technique. Furthermore, we found that traditional rendering techniques, which have often performed poorly in previous studies, showed comparable performance to physically based methods in our study.

Figures (11)

• Figure 1: A screenshot of the user interface during the evaluation. The slider below the rendering was used to estimate the depth of the grey estimation point between the green and purple bounding points.
• Figure 2: Distribution of distances and depths between evaluation points among test cases: relative depth of the estimation point, on-screen (2D) distance between the bounding points, volumetric (3D) distance between the bounding points, and depth difference between the bounding points. The on-screen distance has been computed in normalized device coordinates, and the volumetric distance has been computed such that the volume was a cube with a unit side length.
• Figure 3: Test volumes: mitos (top left), fibers (top right), pebbles (bottom left), and manix (bottom right). The images were rendered with volumetric path tracing.
• Figure 4: The rendering methods used in the experiment: ISO (top left), EAM (top right), DOS (bottom left), and VPT (bottom right).
• Figure 5: Distribution of ages among study par-ti-ci-pants.