Temporally coherent real-time labeling of dynamic scenes

The augmentation of objects by textual annotations provides a powerful means for visual data exploration. Especially in interactive scenarios, where the view on the objects and, thus, the preferred placement of annotations changes continually, efficient labeling procedures are required. As identified by a preliminary study for this paper, these procedures have to consider a number of requirements for achieving an optimal readability, e.g. cartographic principles, visual association and temporal coherence. In this paper, we present a force-based labeling algorithm for 2D and 3D scenes, which can compute the placements of annotations at very high speed and fulfills the identified requirements. The efficient labeling of several hundred annotations is achieved by computing their layout in parallel on the GPU. This allows for a real-time and collision-free arrangement of both dynamically changing and static information. We demonstrate that our method supports a large variety of applications, e.g. geographical information systems, automotive navigation systems, and scientific or information visualization systems. We conclude the paper with an expert study which confirms the enhancements brought by our algorithm with respect to visual association and readability.