AI4CycleMaps - AI-based detection, segmentation, localisation, and geometric interpretation for future cycling infrastructure mapping

Project Overview

AI4CycleMaps is a research initiative focused on the use of artificial intelligence to analyze bicycle infrastructure using image-based data sources. The project investigates how aerial images, street-level images, publicly available open data sources, and supplementary data collected via mobile mapping can be processed using recognition, segmentation, and visual language models to identify relevant infrastructure elements and improve the automation of workflows for interpreting the infrastructure.

Currently, the project is focused on testing AI models and developing initial workflows for the detection, localization, and geometric characterization of objects and surface elements such as bike lanes, red bike lanes, road markings, traffic signs, cyclists, and related features of urban mobility. A key objective is to reduce manual effort and transition to more automated, reproducible, and scalable workflows for analyzing bicycle infrastructure.

Why AI4CycleMaps?

Addressing the challenges of manual and fragmented data collection for bicycle infrastructure: The project originated from the current challenges in collecting, maintaining, and harmonizing data on bicycle infrastructure. In many cases, this information is still collected through on-site inspections, manual evaluations, and institution-specific documentation processes.
High manual effort: Infrastructure inventories often require time-consuming site visits, manual evaluations, and repeated on-site verifications.
Different data structures: Municipalities and institutions may use different formats, schemas, and documentation practices, which complicates integration.
Limited timeliness: Bicycle infrastructure changes over time, but manual update cycles can cause datasets to quickly become outdated.
Challenges in collaboration: Inconsistent and fragmented data can hinder collaboration between municipalities, researchers, planners, and public agencies.

Project Goal

Localization and Characterization: Development of workflows for localizing detected elements and estimating relevant geometric properties such as position, extent, and spatial structure

Preparation for Mapping: Laying the foundation for future GIS-compatible results through strategies for geospatial post-processing, validation, and vectorization.
Detection: Evaluation of AI models for detecting bicycle-related infrastructure and urban mobility objects in various image sources.

Methodology

Image Data Collection: The project investigates the use of aerial imagery, street-level imagery, public image sources, and supplementary mobile mapping data.
AI-Based Analysis: Recognition, segmentation, and image-to-language models are evaluated to detect elements of bicycle infrastructure.
Geodata Processing: The detected elements will be linked to spatial information to enable their localization, geometric characterization, and GIS integration.

Validation and mapping: Future project phases will focus on validation, structured data export, and GIS-compatible representations of bicycle infrastructure.

Data Sources

Street-level imagery: Street-level imagery provides contextual information on signs, markings, lane structure, and the street environment.
Open geodata: Publicly available geodata sets can support candidate selection, spatial referencing, and comparison.

Mobile mapping data: Supplementary mobile mapping data can be used to provide additional visual and spatial information in selected test areas.
Aerial images: Aerial images can support the analysis of bicycle infrastructure from a bird’s-eye view.

Project Members

Junior Professorship in Geosensor Systems, TU Dresden
Prof. apl. Dr. Anette Eltner
Dr.-Ing. Ahmad El-Alailyi
Dr.-Ing. Pedro Alberto Pereira Zamboni
Dr. Lida Asgharian Pournodrati

Chair of Mobility Systems Planning, TU Dresden
Prof. Dr.-Ing. Regine Gerike
Dr.-Ing. Sebastian Hantschel
Dr.-Ing. Armin Kollascheck

Funding

The project runs from December 1, 2025, to November 30, 2028.