Projects urban safety

bikeped

Real-time bicycle and pedestrian safety system and evaluation testbed for urban intersections.

Role
Creator and maintainer
Paper
arXiv preprint, 2026
A 200-degree fisheye view under a viaduct with a cyclist approaching a pedestrian, the alert link between them, the same moment on the ground plane, and the cyclist-to-pedestrian distance over time with the alert interval shaded
A cyclist approaching a pedestrian in a real recording from the 200-degree fisheye camera. Detections are projected to the ground plane with the system's fisheye model, and the bikeped decision pipeline, run with its deployed settings, raises the alert 1.8 s before the closest approach.

bikeped is a real-time pedestrian and cyclist safety system built around wide-angle perception, edge inference, and reproducible scenario testing for urban intersections.

The system combines fisheye calibration, fisheye-aware object detection, ground-plane projection, a decision layer for alerts, and a conformance testbed with paired schematic and simulator videos. The public materials include code, replication scripts, documentation, a simulator, and a companion dataset.

Project details

Artifacts
GitHub implementation, documentation, simulator, conformance scenarios, and Hugging Face dataset
Keywords
  • Bicycle and pedestrian safety
  • Fisheye perception
  • Edge AI
  • Real-time alerts
  • Ground-plane projection
  • CARLA scenarios
  • Conformance testing

References

2026

  1. A Real-Time Bike-Pedestrian Safety System with Wide-Angle Perception and Evaluation Testbed for Urban Intersections
    A Real-Time Bike-Pedestrian Safety System with Wide-Angle Perception and Evaluation Testbed for Urban Intersections
    Mehmet Kerem Turkcan
    Apr 2026