Scholarlink

Traffic anomaly detection is essential for road safety and efficient traffic management. This study presents a customized traffic anomaly dataset with images from web scraping and real-time captures between Palakollu and Bhimavaram, Andhra Pradesh. It classifies images into “Anomaly” (e.g., crashes, potholes) and “No Anomaly” (clear roads). We evaluate four deep learning models MobileNetV2, InceptionV3, DenseNet201, and VGG16 for anomaly detection. The results provide insights for improving real-time traffic monitoring, road safety, and automated traffic management. The models are evaluated by different evaluation parameters. MobileNetV2, and InceptionV3 report a superior accuracy of 94% than rest two models. Moreover, MobileNetV2 exhibits a balanced performance for both anomaly and no-anomaly cases.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Image-Based Traffic Anomaly Detection Using Deep Learning Models

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Srinivasa Rao Tottempudi,
  • Debasis Mohapatra,
  • Nallala Mohana Lakshmi

摘要

Traffic anomaly detection is essential for road safety and efficient traffic management. This study presents a customized traffic anomaly dataset with images from web scraping and real-time captures between Palakollu and Bhimavaram, Andhra Pradesh. It classifies images into “Anomaly” (e.g., crashes, potholes) and “No Anomaly” (clear roads). We evaluate four deep learning models MobileNetV2, InceptionV3, DenseNet201, and VGG16 for anomaly detection. The results provide insights for improving real-time traffic monitoring, road safety, and automated traffic management. The models are evaluated by different evaluation parameters. MobileNetV2, and InceptionV3 report a superior accuracy of 94% than rest two models. Moreover, MobileNetV2 exhibits a balanced performance for both anomaly and no-anomaly cases.