Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Learning Traffic Anomalies from Generative Models on Real-Time Observations

Created by
  • Haebom

Author

Fotis I. Giasemis, Alexandros Sopasakis

Outline

In this paper, we present a Spatiotemporal Generative Adversarial Network (STGAN)-based model that accurately detects traffic anomalies using data from 42 traffic cameras in Gothenburg, Sweden, in 2020. STGAN captures the complex spatial and temporal dependencies of traffic data by combining Graph Neural Networks and Long Short-Term Memory networks. Real-time traffic volume data in units of minutes are converted into flow indicators representing vehicle density, which are used as inputs to the model. The model is trained with data from April to November 2020 and validated with data from November 14 to 23. As a result, we demonstrate that the model effectively detects traffic anomalies (such as camera signal outages, visual artifacts, and extreme weather conditions) with high accuracy and low false positive rate.

Takeaways, Limitations

•
Takeaways:
◦
Empirically demonstrating high accuracy and efficiency of a traffic anomaly detection model using STGAN.
◦
Validation of the effectiveness of spatial-temporal data analysis through the combination of Graph Neural Networks and LSTM networks.
◦
Development and validation of practical models using real traffic data.
◦
Suggests the possibility of detecting various types of traffic anomalies (camera errors, bad weather, etc.).
•
Limitations:
◦
Regional limitations of the data used (Gothenburg, Sweden). Further research is needed to determine generalizability to other regions.
◦
The period used to evaluate the model's performance is relatively short. Long-term performance verification is required.
◦
There is a possibility that the detection performance for certain types of anomalies may be relatively lower than for others. Model improvement and further research are required.
◦
There is potential for improving model performance by considering features other than the ones used (vehicle density).
👍
Made with Slashpage