Authors: Amin Karimi Monsefi, Pouya Shiri, Ahmad Mohammadshirazi, Nastaran Karimi Monsefi, Ron Davies, Sobhan Moosavi, Rajiv Ramnath

Published on: February 07, 2024

Impact Score: 8.2

Arxiv code: Arxiv:2402.05151

Summary

• What is new: The introduction of CrashFormer, a new multi-modal architecture for predicting traffic accidents using a comprehensive yet accessible range of inputs.
• Why this is important: The need for a more generalizable, reproducible, and practically feasible approach to predicting traffic accidents.
• What the research proposes: CrashFormer combines historical accidents, weather data, map images, and demographic info to predict accident risks every six hours in specific areas.
• Results: In 10 major US cities, CrashFormer surpassed existing sequential and non-sequential models in accuracy, with a 1.8% improvement in F1-score using sparse data.

Technical Details

Technological frameworks used: CrashFormer – multi-modal architecture

Models used: Sequential encoder, Image encoder, Raw data encoder, Feature fusion module, Classifier

Data used: History of accidents, weather information, map imagery, demographic info

Potential Impact

Traffic management and urban planning sectors, insurance companies, public safety and regulatory bodies could significantly benefit from or be disrupted by CrashFormer’s insights.

Want to implement this idea in a business?

We have generated a startup concept here: SafeRoutes.