Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ChatGPT as a mapping assistant: A novel method to enrich maps with generative AI and content derived from street-level photographs

Levente Juhász, Peter Mooney, Hartwig H. Hochmair, Boyuan Guan

TL;DR

This paper investigates using generative AI to enrich maps by turning street-level image descriptions into OpenStreetMap road tags, integrating VGI sources (OSM and Mapillary) with off-the-shelf AI tools. The authors combine GPT-3.5-turbo and BLIP-2 in a Miami-area study, evaluating four prompt scenarios (Baseline, Locational context, Object detection, and OD+LC) against human analysts across 94 road segments. Results show that richer image descriptions and additional context substantially improve tagging accuracy, with OD+LC achieving the largest gains (up to ~19.5 percentage points) and semantic road-category tagging reaching up to ~60% accuracy in some configurations. The study highlights the potential and limitations of AI-assisted mapping, suggesting that larger-scale studies and multimodal spatial AI systems could advance practical GIS workflows while mitigating issues like hallucinations and tagging heterogeneity.

Abstract

This paper explores the concept of leveraging generative AI as a mapping assistant for enhancing the efficiency of collaborative mapping. We present results of an experiment that combines multiple sources of volunteered geographic information (VGI) and large language models (LLMs). Three analysts described the content of crowdsourced Mapillary street-level photographs taken along roads in a small test area in Miami, Florida. GPT-3.5-turbo was instructed to suggest the most appropriate tagging for each road in OpenStreetMap (OSM). The study also explores the utilization of BLIP-2, a state-of-the-art multimodal pre-training method as an artificial analyst of street-level photographs in addition to human analysts. Results demonstrate two ways to effectively increase the accuracy of mapping suggestions without modifying the underlying AI models: by (1) providing a more detailed description of source photographs, and (2) combining prompt engineering with additional context (e.g. location and objects detected along a road). The first approach increases the suggestion accuracy by up to 29%, and the second one by up to 20%.

ChatGPT as a mapping assistant: A novel method to enrich maps with generative AI and content derived from street-level photographs

TL;DR

This paper investigates using generative AI to enrich maps by turning street-level image descriptions into OpenStreetMap road tags, integrating VGI sources (OSM and Mapillary) with off-the-shelf AI tools. The authors combine GPT-3.5-turbo and BLIP-2 in a Miami-area study, evaluating four prompt scenarios (Baseline, Locational context, Object detection, and OD+LC) against human analysts across 94 road segments. Results show that richer image descriptions and additional context substantially improve tagging accuracy, with OD+LC achieving the largest gains (up to ~19.5 percentage points) and semantic road-category tagging reaching up to ~60% accuracy in some configurations. The study highlights the potential and limitations of AI-assisted mapping, suggesting that larger-scale studies and multimodal spatial AI systems could advance practical GIS workflows while mitigating issues like hallucinations and tagging heterogeneity.

Abstract

This paper explores the concept of leveraging generative AI as a mapping assistant for enhancing the efficiency of collaborative mapping. We present results of an experiment that combines multiple sources of volunteered geographic information (VGI) and large language models (LLMs). Three analysts described the content of crowdsourced Mapillary street-level photographs taken along roads in a small test area in Miami, Florida. GPT-3.5-turbo was instructed to suggest the most appropriate tagging for each road in OpenStreetMap (OSM). The study also explores the utilization of BLIP-2, a state-of-the-art multimodal pre-training method as an artificial analyst of street-level photographs in addition to human analysts. Results demonstrate two ways to effectively increase the accuracy of mapping suggestions without modifying the underlying AI models: by (1) providing a more detailed description of source photographs, and (2) combining prompt engineering with additional context (e.g. location and objects detected along a road). The first approach increases the suggestion accuracy by up to 29%, and the second one by up to 20%.
Paper Structure (14 sections, 3 figures, 4 tables)

This paper contains 14 sections, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Study setup
  3. Data sources and preparation
  4. Resources
  5. AI tools and models
  6. Analysts
  7. Methodology for suggesting OSM tags
  8. Results
  9. Accuracy of suggesting road categories
  10. Additional tag suggestions
  11. Limitations of the results
  12. Summary and discussion
  13. Data availability
  14. Acknowledgements

Figures (3)

  • Figure 1: OpenStreetMap roads and Mapillary images in the study area near Downtown Miami
  • Figure 2: Workflow for preparing input from Mapillary
  • Figure 3: Workflow of using ChatGPT to suggest OSM "highway" tags