In the intelligent management of city traffic, the object detection of UAV vehicles plays a crucial role. Fixed-view cameras cannot comprehensively monitor urban road traffic, and using the flexibility and wide-area view of UAV can supplement the monitoring range of fixed cameras, improve intelligent traffic management and enhance urban traffic management efficiency. However, the limited resources of the UAV load platform and the unique viewpoint and flexibility of UAV imaging lead to image degradation and uneven sample distribution. To address these problems, a UAV vehicle obejct detection method based on TOOD was proposes in this study, where we use a lightweight network, RegNet-800M, as the feature extraction backbone network to reduce computational overhead and improve feature extraction. At the same time, we incorporate an attention mechanism module based on RegNet-800M to optimize the information processing capability of the model through both spatial and channel dimensions. The results show that the proposed method achieves an average accuracy of 55% and a detection speed of 31.6 FPS on the VisDone-2019 subset, and by balancing accuracy and speed, the method meets real-time standards and practical industry requirements.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    UAV Vehicle Detection Method and Application Based on Improved TOOD


    Contributors:
    Ou, Kaitong (author) / Dong, Chaojun (author) / Huang, Wanxia (author) / Lu, Xiaotian (author)


    Publication date :

    2023-08-18


    Size :

    4986450 byte




    Type of media :

    Conference paper


    Type of material :

    Electronic Resource


    Language :

    English




    Vehicle detection method based on improved YOLOv3

    Qi, Cheng / Shen, Xizhong | IEEE | 2022


    Vehicle Light Detection Method Based on Improved CenterNet

    Li, Shisheng / Mao, Li / Wang, Liming et al. | Springer Verlag | 2022


    Research on Vehicle Detection Method Based on Improved YOLOv5s

    Ma, Liangliang / Zhong, Runlu / Shi, Xiaohong et al. | IEEE | 2024