Analysis of traffic data is an essential component of many intelligent transportation system applications where the quality of data plays an important role. Traffic data collected through sensors such as loop detectors often contain anomalies, e.g. due to malfunctioning detectors or anomalous traffic conditions. Regardless of the rooting cause, such data can heavily affect the results of the subsequent analysis (e.g. traffic prediction). There are several challenges regarding anomaly detection, including absence of universal definition of anomaly, change of traffic pattern over time, as well as unavailability of labeled data, use-case driven analysis. In this paper, a new anomaly detection method for traffic univariate time-series is proposed which does not assume labeled historical data yet uses expert feedback to deal with the fluid definition of anomaly. The method is exemplified and evaluated by applying it on real traffic time series data collected through loop detectors installed in an urban road network in Europe. Employing the proposed method as a pre-process of traffic state estimation can increase the accuracy measure as well as ease the learning of different traffic patterns.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    Adaptable Anomaly Detection in Traffic Flow Time Series


    Beteiligte:


    Erscheinungsdatum :

    2019-06-01


    Format / Umfang :

    834641 byte




    Medientyp :

    Aufsatz (Konferenz)


    Format :

    Elektronische Ressource


    Sprache :

    Englisch




    Traffic Anomaly Detection Via Conditional Normalizing Flow

    Kang, Zhuangwei / Mukhopadhyay, Ayan / Gokhale, Aniruddha et al. | IEEE | 2022



    Anomaly Scoring for Prediction-Based Anomaly Detection in Time Series

    Li, Tianyu / Comer, Mary L. / Delp, Edward J. et al. | IEEE | 2020


    Anomaly detection in time series via robust PCA

    Jin, Yongjun / Qiu, Chenlu / Sun, Lei et al. | IEEE | 2017