Promoting cycling is a crucial solution to improve the livability of urban environments. A non-invasive way to promote cycling is to enhance the bike ride experience via Green Light Optimal Speed Advisory (bike-GLOSA). This work presents such a system for Hamburg, making bike-GLOSA practical in real-world urban environments and evaluating its impact on bike rides.
GLOSA systems ingest data from traffic lights to predict their switching behavior and provide speed advisories to users. In the first challenge, we need to find whether the data is reliable, overseeing thousands of traffic lights throughout the city area. We develop a quality assurance framework and monitor the prediction quality over a longer period. Affected by data outages, we achieve a median prediction availability of 55% (IQR: 28%) and prediction quality of 86%. We identify concrete weak points, enhancing prediction stability by 11% to 66% (IQR: 17%) measured in 2024. Furthermore, based on four weeks of data for 18,009 traffic lights, we find that traffic adaptivity may be less problematic for traffic light prediction than previously envisioned.
Afterward, we develop a novel method to match traffic lights along the user's trajectory. Instead of using the user's location or camera, our method matches traffic lights along a precalculated bike route geometry. However, errors and inaccuracies in bike routes challenge this approach, requiring an advanced model that employs spatial reasoning to find which associated traffic light turn geometries match the given route geometry. The final model achieves matching F1 scores of 92% and 86% validated on a separate dataset, requiring a median extra time of 1.4 seconds during bike route calculation.
Building on these results, we focus on reducing bike routing errors and enhancing route alignment with actual bike paths. Our solution involves an authoritative bike routing dataset and cross-border routing to OpenStreetMap. Apart from a more consistent surface coverage, we find better alignment with actual cycling infrastructure, traffic lights, and user trajectories than with other routing providers. Enhanced alignment and 37% fewer routing errors lead to a 4.74% increase in traffic light matching F1 score. A route-based distance-to-signal estimation is proposed, showing a more stable distance estimation than the over-the-air distance from related work.
We combine the developed components in a smartphone app and conduct an unsupervised long-term test throughout 2023 with Hamburg citizens. Survey responses suggest a twofold effect of the speed advisory: rolling out in anticipation of red and accelerating to catch green. These effects are also visible in the recorded data. Approaches with adherence to the speed advisory have 15.32% fewer stops but a 3.3% increase in energy expenditure to catch the green phase by cycling 2.92 km/h faster. Approaches without adherence have a 12.85% higher chance of stopping and a 3.39 km/h decrease in speed while saving 5.5% in energy and rolling out earlier. Combined, these cases cancel out each other, with a 0.74 km/h slower traffic light passing speed, 1.4% estimated energy savings, 0.73% increased stop rate, and 1.46 seconds increased waiting time when stopped.
Based on the survey, users report a System Usability Scale of 73, with improvable reliability and coverage of speed advisories. Among many ways to improve our developed solution, users see enhanced informedness, reduced stops, and increased comfort as key benefits. We thoroughly analyze these findings and outline potential directions for future research.
Identifer | oai:union.ndltd.org:DRESDEN/oai:qucosa:de:qucosa:92544 |
Date | 19 July 2024 |
Creators | Matthes, Philipp |
Contributors | Sommer, Christoph, Sester, Monika, Technische Universität Dresden |
Source Sets | Hochschulschriftenserver (HSSS) der SLUB Dresden |
Language | English |
Detected Language | English |
Type | info:eu-repo/semantics/publishedVersion, doc-type:doctoralThesis, info:eu-repo/semantics/doctoralThesis, doc-type:Text |
Rights | info:eu-repo/semantics/openAccess |
Page generated in 0.0019 seconds