Speaker: Florian Wagner
Abstract
Placing labels on interactive 3D maps is challenging: traditional methods produce high-quality results but are too slow for real-time interaction, while fast heuristic approaches often create overlapping or poorly positioned labels. The problem becomes even harder in 3D terrain visualization, where elevation differences cause occlusions, camera movements require constant label adjustments, and zoom levels change how labels should appear and scale.
This thesis proposes a new approach that combines CPU and GPU processing to achieve both quality and speed. The system uses the GPU to evaluate many possible label positions in parallel, checking for terrain visibility and overlaps using a grid-based method. The CPU then selects the best placements and applies temporal coherence filtering—a particularly challenging aspect, as labels must remain stable during camera movements without distracting flickering or jumping, while still adapting to the changing 3D view.
The goal is to enable smooth, interactive label placement in 3D map visualizations without sacrificing cartographic quality, addressing terrain-specific challenges that existing 2D or simplified 3D approaches don't handle well.