Vehicle behavior recognition is a critical yet challenging issue in supporting the next generation of intelligent transportation systems (ITSs). To achieve accurate and effective vehicle behavior recognition, this work presents a fine-grained recognition framework named Struct Segment Temporal Convolution Network (SS-TCN), which enables the recognition of different types of vehicle behaviors via multi-sensor fusion. Specifically, we first present a three-stage vehicle behavior proposal scheme, in which the vehicle behavior is modeled by the starting, coursing, and ending stages, and the sensor sequence data of the corresponding stages are input into the neural network to capture contextual information of vehicle behaviors. Second, we propose a Sensor Channel-wised Attention Module (SCAM), which is embedded into the SS-TCN to learn sensor weights for different vehicle behaviors. Furthermore, to adapt the temporal variation of vehicle behaviors, we design a Stage-wised Temporal Pyramid Pooling (STPP) method to convert input features with variable length into fixed-length representation, ensuring effective training and inference of SS-TCN. Finally, we conduct experiments on two vehicle behavior datasets and the results conclusively demonstrate the feasibility and efficiency of the proposed framework and corresponding solutions.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    A Fine-Grained Vehicle Behavior Recognition Framework: Struct Segment Temporal Convolutional Network


    Beteiligte:
    Yan, Guozhi (Autor:in) / Liu, Kai (Autor:in) / Hu, Junbo (Autor:in) / Jin, Feiyu (Autor:in) / Zhang, Hao (Autor:in)


    Erscheinungsdatum :

    2023-09-24


    Format / Umfang :

    633463 byte





    Medientyp :

    Aufsatz (Konferenz)


    Format :

    Elektronische Ressource


    Sprache :

    Englisch







    Dual-Rank Attention Module for Fine-Grained Vehicle Model Recognition

    Cai, Wen / Zhu, Wenjia / Cheng, Bo et al. | British Library Conference Proceedings | 2022