TurnsMap: Enhancing Traffic Safety with Crowdsensing and Deep Learning

Abstract

Left turns across oncoming traffic are among the most dangerous driving maneuvers, responsible for a disproportionate share of intersection fatalities. Yet navigation apps provide no safety guidance for turns — they only tell you when to turn, not whether it is safe to do so.

TurnsMap is a crowdsensing and deep learning framework that builds a safety map of intersections from smartphone sensor data collected during everyday driving. By analyzing the motion patterns of drivers who have made the same turn, TurnsMap learns which intersections are consistently dangerous and alerts future drivers before they encounter them — democratizing traffic safety information for navigation apps.

Key Contributions

• First crowdsensing system to map intersection-level turn safety from commodity smartphones.
• Deep learning model distinguishes safe and dangerous left turns from motion sensor data.
• No dedicated infrastructure — data collected passively from everyday driving.
• Compatible with existing navigation apps as a safety overlay layer.

BibTeX

@article{chen2019turnsmap,
  title     = {TurnsMap: Enhancing Traffic Safety with Crowdsensing and Deep Learning},
  author    = {Chen, Dongyao and Shin, Kang G.},
  journal   = {Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies (IMWUT)},
  year      = {2019},
  publisher = {ACM}
}