Fehler, Richard (FZI Research Center for Information Technology), Rösch, Kevin (FZI Forschungszentrum Informatik), Immel, Fabian (FZI Research Center for Information Technology), Stiller, Christoph (Karlsruhe Institute of Technology)

Robust Traffic Light Detection and Stopline Release by HD-Map Association

Scheduled for presentation during the Regular Session "S05c-Deployment, Modeling, and Optimization in Intelligent Transportation Systems" (WE-LA-T5), Wednesday, November 19, 2025, 16:00−16:20, Surfers Paradise 2

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 19, 2025

Abstract

We present a robust system architecture for perceiving traffic lights and determining stop line release states for automated vehicles. Unlike traditional approaches that rely on error-prone back-projection of mapped traffic lights, our method employs deep learning-based detection and classification coupled with a Min-Cost-Flow association algorithm to match detected traffic lights with mapped traffic lights. This association result subsequently links back to the relationship stored in the map between multiple traffic lights and the stop line relevant for the ego-vehicle, improving the robustness of stop line release state estimation against localization, calibration, and mapping errors. Our modular pipeline includes detection, classification, 3D extension, matching and temporal filtering, handling imperfect perception. Evaluation during closed-loop autonomous operation on public roads in two cities of Germany demonstrates 96.68% accuracy, estimating the release state across 211 stop line traversals under diverse conditions including rain and low-light scenarios.