Modeling How Humans Judge Dot-Label Relations in Point Cloud Visualizations.

Area of Science:

• Information Visualization
• Human-Computer Interaction
• Cognitive Psychology

Background:

• Current point cloud labeling methods lack established guidelines, unlike cartography.
• Naive labeling strategies can lead to misinterpretations of point-to-label associations.

Purpose of the Study:

• To investigate factors influencing human interpretation of labeled point clouds.
• To develop and evaluate predictive models for human judgment in point cloud labeling.
• To inform the creation of improved labeling strategies for information visualization.

Main Methods:

• Derived a typology of labeled point cloud representations from existing applications.
• Defined computational models based on geometric properties to predict human interpretation.
• Conducted an empirical user study to assess model accuracy in predicting human judgments.

Main Results:

• Point cluster presence, label size, and viewing angle significantly affect user judgment.
• Different distance metrics perform variably, suggesting label centers are unreliable reference points.
• The developed models showed predictive capability for human interpretation.

Conclusions:

• Geometric properties like clusters, size, and angle are crucial for labeled point cloud perception.
• Relying solely on label centers for association is discouraged due to metric performance variations.
• Findings provide a foundation for developing more effective and less misleading point cloud labeling techniques.