期刊目次

加入编委

期刊订阅

添加您的邮件地址以接收即将发行期刊数据:

Open Access Article

Scientific Development Research . 2024; 4: (1) ; 51-56 ; DOI: 10.12208/j.sdr.20240010.

Research on autonomous obstacle avoidance algorithm of UAV based on deep learning
基于深度学习的无人机自主避障算法研究

作者: 何庆新 *

闽南理工学院 福建石狮

*通讯作者: 何庆新,单位:闽南理工学院 福建石狮;

发布时间: 2024-12-18 总浏览量: 293

摘要

本研究针对无人机自主避障算法进行研究,选用深度学习领域的先进模型YOLOv10作为核心检测算法,旨在提升无人机在复杂环境中的避障效率和飞行安全性。结合双目视觉测距技术,实现了基于YOLOv10的障碍物精准检测与三维定位,同时推导了坐标转换关系及障碍物三维坐标的最优近似解。算法设计涵盖图像预处理、模型推理、障碍物检测和避障策略制定等关键步骤。 实验选用Open Image数据集和自定义图像,在多种光照、遮挡和速度条件下对YOLOv10进行了性能评估。结果显示,YOLOv10在无人机避障任务中表现出高准确率、高避障成功率及良好的光照和遮挡适应性,且在高速飞行时性能稳定。本研究为相关领域的算法优化提供了新的思路和方法,具有重要的理论和实践意义。

关键词: 无人机自主避障;YOLOv10算法;深度学习;双目视觉测距

Abstract

This research focuses on the autonomous obstacle avoidance algorithm of unmanned aerial vehicles (UAVs). The advanced model YOLOv10 in the field of deep learning is selected as the core detection algorithm, aiming to improve the obstacle avoidance efficiency and flight safety of UAVs in complex environments. Combined with the technology of binocular vision ranging, the precise detection and three-dimensional positioning of obstacles based on YOLOv10 are achieved. At the same time, the coordinate transformation relationship and the optimal approximate solution of the three-dimensional coordinate of obstacles are derived. The algorithm design covers key steps such as image pre-processing, model inference, obstacle detection, and obstacle-avoidance strategy formulation. Experiments use the Open Image dataset and custom images to evaluate the performance of YOLOv10 under the condition of various lighting, blocking, and speeding. The results show that YOLOv10 exhibits high accuracy, high obstacle-avoidance success rate, and good adaptability to lighting and blocking in UAV obstacle-avoidance tasks, and its performance is stable during high-speed flight. This research provides new ideas and methods for algorithm optimization in related fields and has important theoretical and practical significance.

Key words: Autonomous obstacle avoidance of UAVs; YOLOv10; Deep learning; Binocular vision ranging

参考文献 References

[1] 何晨阳,王耀力,常青,等.基于预探索的动态自适应3DVFH+无人机避障算法[J].电子设计工程,2024,32(04):1-7.

[2] 刘玄冰,周绍磊,肖支才,等.无人机避障方法研究综述[J].兵器装备工程学报,2022,43(05):40-47.

[3] 张宏宏,甘旭升,毛亿,等.无人机避障算法综述[J].航空兵器,2021,28(05):53-63.

[4] 曹益荣.无人机避障飞行系统关键技术的研究与实现[D].成都:电子科技大学,2021.

[5] 王义和.基于深度学习的无人机智能避障系统设计与实现[J].信息与电脑(理论版),2024,36(14):108-110

[6] 岳颖搏.基于深度学习的无人机障碍物检测与避障方法研究[D].西安:西安工业大学,2024.

[7] 王金云.基于深度学习的无人机电力巡检故障检测与避障系统[D].天津:天津理工大学,2023.

[8] 常可铮,李耀成,杨得武.基于边缘信息的轻量化图像去雾卷积神经网络[J].信息技术与信息化,2024,(09):63-66.

[9] 赵程栋,庄继晖,程晓鸣,等.基于特征注意力机制的RNN-Bi-LSTM船舶轨迹预测[J].广东海洋大学学报,2022,42(05):102-109.

[10] 申静,曾晴,高雅楠,等.基于YOLOv10s的车辆与行人检测方法[J].电脑知识与技术,2024,20(24):25-27+39.

[11] 黄毅,周纯,刘欣军,等.基于YOLOv10的无人机复杂背景下多尺度检测模型[J/OL].光通信研究,1-8[2024-10-24].

引用本文

何庆新, 基于深度学习的无人机自主避障算法研究[J]. 科学发展研究, 2024; 4: (1) : 51-56.