In recent years, with the rise of the application of deep learning tasks in the end of multi-rotor UAVs, it is a very scientifically meaningful and practical research to combine artificial intelligence technology with pattern recognition technology to realize automatic target detection and autonomous obstacle avoidance in UAVs. The traditional manual inspection method has the problems of low efficiency, high labor and material consumption and low safety, so the country is vigorously developing the drone industry at this stage. UAV inspection tasks are becoming more and more important, and it has become a trend for UAVs to shift to the direction of unmanned operations. With the increase of UAV density in large cities, it is meaningful to control and manage the trajectory of UAVs and provide intelligent trajectory optimization and anti-collision system. In order to achieve safe and fast autonomous flight, it is necessary to intelligently optimize the flight trajectory of UAV to ensure the best response time and increase the duration of continuous flight. At the same time, due to the low-slow-small characteristics of UAVs and the complexity of cities, UAVs are often faced with the interference of airborne obstacles, such as utility poles or flying birds, and need to perform real-time environment sensing and change their trajectories in time to avoid these obstacles. We propose an ED-YOLO network model for target detection of UAV obstacles. The target detection algorithm is based on YOLOv4, and firstly, a channel attention mechanism is added to the backbone network to improve the detection accuracy without increasing the computation. Compared with YOLOv4, its average accuracy mean of obstacle target detection is only reduced by 1.4%, while the model volume is reduced by 94.9%, the floating-point operation volume is reduced by 82.1%, and the prediction speed is improved by 2.3 times. We also studied UAV obstacle avoidance path planning. Firstly, we propose the UAV obstacle avoidance strategy, secondly, we introduce the traditional artificial potential field method and analyze the defects of this algorithm, and then we propose a solution and conduct simulation and comparison tests in MATLAB software to illustrate its effectiveness.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    UAV Target Detection and Path Planning Based on Improved YOLOv4 Networks with Improved Artificial Potential Field Methods


    Beteiligte:
    Liu, Zhanhan (Autor:in)


    Erscheinungsdatum :

    2023-11-25


    Format / Umfang :

    494131 byte




    Medientyp :

    Aufsatz (Konferenz)


    Format :

    Elektronische Ressource


    Sprache :

    Englisch



    UAV Target Detection Algorithm Based on Improved YOLOv4

    Wang, Wenyue / Li, Jian / Zhang, Qi | Springer Verlag | 2022


    Pedestrian Detection Algorithm Based on Improved YOLOv4

    Zheng, Ziheng / Ni, Chenhui / Zeng, Guolei | IEEE | 2023


    Detection Algorithm Based on Improved YOLOv4 for Pedestrian

    Kong, Wen / Qiao, Yidan / Wei, Ziqi | IEEE | 2022


    UAV Path Planning Based on Improved Artificial Potential Field Method

    Wang, Hai / Wang, Lei / Gao, Xiaohua et al. | Springer Verlag | 2023


    Maritime ship detection algorithm based on improved YOLOv4

    Chen, Wangcheng / Li, Yingshu / Wang, Xuemei et al. | SPIE | 2023