The continuous evolution of smart transportation has led to the vigorous development of research in the field of intelligent vehicle anomaly detection. This study proposes a new traffic incident detection algorithm based on deep learning and combines it with statistical optimization methods for real-time traffic incident monitoring. This model first uses the pre-trained YOLOv5m_c model to detect vehicles and outputs images with vehicle frames from the video stream. At the same time, from the data analysis level, it determines whether there is abnormal behavior between vehicles based on the characteristics of the vehicle frame. The proposed algorithm improves the effect of abnormal events by optimizing the network structure of the YOLOv5 model. Then, starting from the sensor level, the detection method based on the Internet of Vehicles collects and shares vehicle status information in real time. This analysis process fully considers the task offloading and migration strategy in the edge computing environment. Quantitative experimental results show that the proposed algorithm can effectively improve the detection accuracy of traffic incidents and reduce the false alarm rate in real-time traffic data.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    Traffic Incident Detection Algorithm based on Deep Learning: A Statistical Optimization Approach


    Contributors:
    Jing, Zhao (author)


    Publication date :

    2025-02-11


    Size :

    791871 byte




    Type of media :

    Conference paper


    Type of material :

    Electronic Resource


    Language :

    English



    A Deep Learning Approach for Traffic Incident Detection in Urban Networks

    Zhu, Lin / Guo, Fangce / Krishnan, Rajesh et al. | IEEE | 2018



    Traffic-Incident Detection-Algorithm Based on Nonparametric Regression

    Tang, S. / Gao, H. / IEEE | British Library Conference Proceedings | 2005



    Real-time traffic incident detection based on a hybrid deep learning model

    Li, Linchao / Lin, Yi / Du, Bowen et al. | Taylor & Francis Verlag | 2022