TransitNet2Tensor: Bidirectional Rasterization of GTFS Transit Networks for GeoAI

Institut

Lehrstuhl für Fahrzeugtechnik

Typ

Bachelorarbeit / Semesterarbeit / Masterarbeit /

Inhalt

experimentell / theoretisch / konstruktiv /  

Beschreibung

Public transport data is usually published as GTFS feeds containing stops, routes, and trip shapes. Modern
spatial ML and GeoAI models, however, typically work with fixed-size raster grids rather than irregular
graphs. This thesis develops an end-to-end pipeline that takes a real bus network and:
1. Converts it into a simple, compact 4-layer grid representation (stop density, line intensity, direction
encoded as sin/cos), and 2. Reconstructs a meaningful network back from that grid using classical algorithms
or ML models, like encoder-decoder Networks.
The core idea is to build something concrete and usable: a tool that can rasterize a full city’s bus network
into images, and then recover stops, corridors, and connections from those images with measurable fidelity.
Students will work with real GTFS data from multiple cities and build a clean, reproducible pipeline that can
serve as a baseline for future GeoAI research.

 

Work Packages
• Study how GTFS data describes bus networks and how raster grids are typically used in spatial ML.
• Define a clear 4-layer tensor format (gridsize, spatial extent, flow definition, normalization).
• Implement a deterministic rasterization pipeline (stops, line geometry, and directional encoding).
• Implement at least one reconstruction pipeline: thresholding + skeletonization + junction detection
+ stop snapping + corridor extraction.
• Optionally implement a second reconstruction method (e.g. streamline tracing or a light UNet that
predicts node/edge likelihoods).
• Evaluate reconstruction accuracy (stop recovery, corridor similarity, connectivity, robustness) across
cities and resolutions.
• Deliver a reusable codebase and documented examples for future students and researchers.

 

 

Voraussetzungen

Prerequisits
• Programming skills in Python
• Structured and independent working style

Tags

FTM Studienarbeit, FTM SM, FTM Zacher, FTM Informatik

Möglicher Beginn

sofort

Kontakt

Till Zacher, M.Sc.
Tel.: +49 89 289 15351
till.zachertum.de

Ausschreibung