Evaluating label placement for augmented reality view management

Abstract

View management, a relatively new area of research in Augmented Reality (AR) applications, is about the spatial layout of 2D virtual annotations in the view plane. This paper represents the first study in an actual AR application of a specific view management task: evaluating the placement of 2D virtual labels that identify information about real counterparts. Here, we objectively evaluated four different placement algorithms, including a novel algorithm for placement based on identifying existing clusters. The evaluation included both a statistical analysis of traditional metrics (e.g. counting overlaps) and an empirical user study guided by principles from human cognition. The numerical analysis of the three real-time algorithms revealed that our new cluster-based method recorded the best average placement accuracy while requiring only relatively moderate computation time. Measures of objective readability from the user study demonstrated that in practice, human subjects were able to read labels fastest with the algorithms that most quickly prevented overlap, even if placement wasn't ideal.