object tracking is one of the most sophisticated and least researched tasks in computer vision, especially with respect to unmanned aerial vehicles. Primarily it caused by several challenges such as high distance to the tracking objects, variety in object sizes, camera motion, etc. This paper demonstrates the implementation of a tracking pipeline using the YOLOv4-based architecture network defined as YOLOv4eff with the doubled long-short time memory (LSTM) network for road-objects tracking. Our optimization method within YOLOv4eff addresses higher accuracy by improving the network architecture, using modified CSP techniques, Swish activation function, etc. Our object tracker used convolutional network as feature map extractor network of differential sequential frames, YOLOv4eff as detector and doubled long-short time memory (LSTM) network as tracking locations predictor. Our extensive experimental results on self-collected dataset at the height of 10–30 meters and performance comparison with other state-of-the-art tracking methods show that our LYOLOv4eff is more accurate and effectively applicable to objects tracking from a drone.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    Efficient objects tracking from an unmanned aerial vehicle


    Contributors:


    Publication date :

    2021-06-23


    Size :

    731947 byte





    Type of media :

    Conference paper


    Type of material :

    Electronic Resource


    Language :

    English




    Unmanned aerial vehicle target tracking method and system, unmanned aerial vehicle holder and unmanned aerial vehicle

    HU HUAZHI / HE WEIXIONG / HU HAISHENG | European Patent Office | 2022

    Free access


    Efficient Road Detection and Tracking for Unmanned Aerial Vehicle

    Zhou, Hailing / Kong, Hui / Wei, Lei et al. | IEEE | 2015


    Pedestrian tracking from an unmanned aerial vehicle

    Bian, Chao / Yang, Zhen / Zhang, Tao et al. | IEEE | 2016