Detecting crossing pedestrians in complicated traffic situations is a critical problem for intelligent and connected cars (ICVs) in order to minimize traffic accidents. To address this issue, an improved model based on YOLOv5 named Alpha-YOLOv5 is proposed to detect the crossing pedestrian. With the aim of enhancing pedestrian detection performance in crowded and occluded scenes, the native BottleneckCSP module is replaced by the C3TR module, using the TransformerBlock. The Nearest Upsample is replaced with CARAFE to increase global upsampling and training efficiency. In order to address the issue of accurately locating targets in complex traffic scenes, the ESE attention mechanism module is fused in the YOLOv5 model, which improves the model’s ability to capture local information and increases the receptive field. In addition, the Alpha IoU loss is employed to improve the loss and regression accuracy of high IoU targets. Experimental results show that the proposed Alpha-YOLOv5s model outperforms the original YOLOv5 model by achieving a 3.32% increase in mAP for crossing pedestrian detection in real traffic scenes, with an mAP of 95.12%.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    Improved Crossing Pedestrian Detection Model for Intelligent Vehicles in Complex Traffic Scenes


    Contributors:
    Yin, Tianrui (author) / Chen, Wei (author) / Du, Luyao (author) / Xiao, Zhihao (author) / Tong, Bingming (author) / Su, Zixu (author)


    Publication date :

    2023-08-04


    Size :

    1071637 byte





    Type of media :

    Conference paper


    Type of material :

    Electronic Resource


    Language :

    English



    Understanding Pedestrian Behavior in Complex Traffic Scenes

    Rasouli, Amir / Kotseruba, Iuliia / Tsotsos, John K. | IEEE | 2018




    Self-adaptive intelligent pedestrian crossing traffic light system

    YU LAN / LUO ZHONGLI | European Patent Office | 2022

    Free access

    Pedestrian time-to-arrival judgment of vehicles in naturalistic traffic scenes

    Zhao, Qingxian / Liu, Weiren / Zhuang, Xiangling et al. | IEEE | 2021