Source-linked AI summary
From Anonymous Shapes to Named Places: A Tool for Braille and Place-Semantic Annotation of Tactile Maps
Li Liu, Ashmita Dua, Jiaming Qu, David T. Lee, Leilani H. Gilpin
TL;DR
Unlabeled tactile-map footprints are difficult to identify by touch, motivating a tool that adds reader-tailored Braille after 3D geometry generation. The browser workflow matches shapes to OpenStreetMap places, abbreviates names, and checks print-safe dots; formative sessions involved five maps and ten BLV readers, and the authors treat Braille as one output among several.
Problem
Unlabeled tactile maps present readers with anonymous shapes, so touch does not identify which footprint represents a particular place.
Method
A browser-based tool lets a sighted helper match existing 3D-map shapes to OpenStreetMap places, edit abbreviations, and export print-safe Braille plus reusable per-label records.
Results
The tool was demonstrated on five printed maps, and ten BLV readers compared unlabeled and annotated prints; four read Braille fluently.
Takeaways & Limitations
Linking coordinates, printed geometry, and place semantics supports Braille labels and future audio or non-Braille outputs from the same annotations.
Takeaways & Limitations
The tool still requires a sighted annotator, while formative evaluation and Braille support remain limited by abbreviation, contracted-Braille, sizing, and legend issues.
Abstract
from arXiv · showhide
On a 3D-printed tactile map, a building felt under the finger is an anonymous shape: touch alone cannot tell which footprint is which, and a spoken description cannot reliably point to one shape at one place. We present a web-based tool that lets a sighted helper click to add on-shape Braille labels to an already-generated map model, downstream of the geometry generator so that whoever knows the reader and the local Braille standard does the labeling. The tool offers click-based OpenStreetMap matching, hand-editable abbreviation that shrinks a name to fit a footprint, and print-safe dot geometry with a review step that catches anomalies before printing. We demonstrate it on five printed maps of different place types, from a downtown core to a college campus and a small dining mall. In formative sessions in which ten BLV readers compared an unlabeled print with an annotated one, four read Braille fluently, so we treat Braille as one output among several rather than the only one. The tool's core is the link between coordinates, geometry, and a place's semantics, which can drive an audio readout or a non-Braille code. The tool is available at https://leolee7.github.io/Annotate_Braille/.
1 Introduction
Tactile maps can provide spatial knowledge, but unlabeled printed footprints remain anonymous to touch. The paper presents downstream annotation that adds reader-tailored, on-shape Braille and reusable place semantics.
- Problem: Unlabeled tactile maps leave readers unable to identify which printed footprint is the library, station, or other place.The missing layer is a readable, finger-locatable identity on each shape.
- Gap: Existing pipelines often omit labels, automate geometry instead, or target swell paper, leaving complex cases and reader-specific naming to people.The paper identifies a gap after 3D geometry has already been generated.
- Approach: The tool lets a sighted helper annotate an existing 3D-print model using click-based OpenStreetMap matching and hand-editable abbreviations tailored to the reader and local Braille standard.Labeling is deliberately kept downstream of geometry generation.
- Implication: The per-label record links each printed shape to its OpenStreetMap semantics, enabling outputs beyond Braille.The paper specifically points to audio readout and non-Braille forms as later uses.
- Approach: Print-safe Braille geometry and review checks flag anomalies before printing, including dots that pierce the base or labels that overlap or float.The exported record separates print-specific geometry from reusable place semantics.
2 Related Work
Prior work addresses tactile geometry, label placement, device-mediated reading, or Braille legibility separately. This paper positions its contribution as annotating already-generated 3D geometry while preserving reusable place semantics.
- Tactile map generation and labeling: Tactile cartography defines touch-oriented symbols and readable detail density, while readers also need titles, overviews, legends, and distinguishable icons.These requirements become harder as map scale shrinks.
- Tactile map generation and labeling: Most generation tools separate geometry from text: Touch Mapper provides printable geometry without Braille, while computer vision targets tactile features rather than labels.A rule-based 3D-print pipeline exists, but fixed generator rules cannot adapt labels to individual readers or Braille standards.
- Tactile map generation and labeling: Automatic label placement can abbreviate names for swell-paper maps, but complex cases still require human inspection and editing.This supports retaining human judgment in the labeling workflow.
- Device- and agent-based reading aids: Device- and agent-based systems support richer queries, but shape identity resides in companion hardware or input rather than on the printed shape.The paper instead places identity on the artifact through on-shape Braille.
- Braille on a printed surface: Braille symbol research and standards specify legibility and physical dot ranges, but leave open how to place labels on map footprints.The paper combines print-safe on-shape Braille with placement on existing 3D geometry.
3 System
The browser-based system links a 3D STL, OpenStreetMap entities, and a sighted annotator so labels can be matched, edited, checked, and exported without coordinate measurement.
- Annotation workflow: The browser editor displays the 3D model beside a 2D OpenStreetMap view, making rooftops and building polygons clickable for direct pairing.The annotator can select either side first, avoiding manual coordinate measurement.
- Annotation workflow: After pairing, the tool fills the place name, description, Braille code, and feature metadata, while the helper can adjust placement and reopen labels for editing.The workflow supports annotation of an already-generated STL.
- Abbreviation: Hand-editable abbreviations use function keywords, campus codes, or filtered initials, with the full expansion stored in an exported legend key.Examples include Advanced Technology Center → ATC, Building S5 → S5, and Neiman Marcus → NM.
- Print-safe output: Braille dots use hemispheres on flat discs with an adaptive base slab maintaining at least 1 mm of material beneath each dot.Cell and dot dimensions remain within ISO 17049 ranges, and Review flags piercing, overlap, or floating problems.
- Reusable record: The tool exports an annotated STL and a per-label record that separates print geometry from reusable semantics, allowing records to be revised or recomposed across sheets.The page runs entirely in the browser without installation or an account.
4 Workflow and Demonstration
The workflow links a generated tactile-map model to place identities through helper-driven annotation, then reviews and exports reusable label records. Feedback covered five map types and ten BLV readers comparing unlabeled and annotated prints.
- Workflow: A sighted helper loads a generated STL, pairs 2D footprints with 3D shapes, and fills in each place name and Braille code.The linked views remove the need to measure coordinates or compare against a separate static map.
- Workflow: The helper adjusts labels, reviews dots that might pierce the base or collide with neighbors, and exports the annotated model.
- Workflow: The exported per-label record links an OpenStreetMap entity, print position, and Braille, separating reusable meaning from print-specific geometry.It can drive an exported legend or a later press-to-hear readout.
- Feedback: Five printed maps represented varied place types, while three uninvolved research assistants also used the tool before feedback sessions with ten BLV readers.Readers compared unlabeled and annotated prints of the same area and discussed legibility, spacing, and usability.
- Feedback: Four of ten readers read Braille fluently; other readers indicated that labels should also support a key, audio, or non-Braille form.Readers generally found the dots legible, but codes were not always self-explaining.
5 Discussion and Future Work
The tool’s reusable place-linked records support outputs beyond printed Braille, while current use remains constrained by annotation, Braille, and formative-evaluation limitations.
- Limitations: The tool still requires a sighted annotator to confirm place matches and abbreviations.
- Limitations: Current Braille support could improve abbreviation, contracted Braille, cell sizing, and legends for ambiguous codes.One reader who read the codes easily still requested slightly larger labels.
- Limitations: The evaluation is formative, and future work should involve teachers and assistive-technology professionals who would perform annotation.
- Future work: The same per-label records could drive press-to-hear audio or compact non-Braille symbols, including maps kept for newcomers at services centers.
- Future work: Because records use OpenStreetMap entities, annotations could accumulate into a shared layer and ground conversational agents in places under a reader’s finger.
- Conclusion: The paper concludes that downstream labeling and linked per-label records set up audio, symbol, and agent outputs beyond Braille.
6 GenAI Usage Disclosure
The tool uses real OpenStreetMap data and deterministic rules rather than generative AI for labels, Braille codes, abbreviations, and print-safe geometry.
- The tool does not rely on generative AI at run time; its labels come from real OpenStreetMap data.
- Deterministic rules produce the Braille codes, abbreviations, and print-safe geometry.
- The authors reviewed all AI-assisted output and take responsibility for the content, design, ideas, and results.