Traffic sign detection and recognition is a key technology for intelligent assisted driving. This paper proposes a key technology for traffic sign detection and recognition based on improved YOLO_V4 in complex urban scenes. The technology uses K-means++ clustering to generate an anchor that conforms to traffic signs; designs a CSPDarknet33 backbone network that greatly reduces the number of weight calculations optimizing the structure of the YOLO_V4 neural network; uses Focal Loss and balanced cross entropy function to optimize the loss function of YOLO_V4, and completes the traffic sign detection work. Then this paper designs an improved VGG16 neural network, which is used to train three models that recognize warning signs, prohibition signs and indication signs. Finally, according to the results of traffic sign detection, the traffic sign recognition work is further completed, the mAP is 94.48% and the FPS is 27, its detection performance is significantly higher than other networks.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    Multi-scale traffic sign detection algorithm based on improved YOLO_V4


    Contributors:
    Li, Sihan (author) / Cheng, Xin (author) / Zhou, Zhou (author) / Zhao, Ben (author) / Li, Shaoqian (author) / Zhou, Jingmei (author)


    Publication date :

    2022-10-08


    Size :

    2550224 byte




    Type of media :

    Conference paper


    Type of material :

    Electronic Resource


    Language :

    English



    Traffic Sign Detection Algorithm Based on Improved YOLOv7

    Wang, Haodong / Zhang, Xindong | IEEE | 2024


    Multi-scale traffic sign detection model with attention

    Fan, Bei Bei / Yang, He | SAGE Publications | 2021


    Scale-variant traffic sign detection

    Wang, Sijuan / You, Zhiqiang | SPIE | 2019


    Traffic Sign Recognition Algorithm Based on Improved YOLOv5

    Sang, Zhengxiao / Xia, Fuming / Huang, Han et al. | IEEE | 2022


    Traffic sign recognition algorithm based on improved ResNet18

    Hu, Yixin / Ye, Qingyang / Zhu, Xuanqi et al. | SPIE | 2024