北京航空航天大学学报 ›› 2022, Vol. 48 ›› Issue (4): 586-594.doi: 10.13700/j.bh.1001-5965.2020.0613

• 论文 • 上一篇    下一篇

电力场景下基于无人机视觉的运动目标追踪方法

冯雪, 杜猛俊, 向新宇, 钱锦, 张敏   

  1. 国网浙江省电力有限公司杭州供电公司, 杭州 310000
  • 收稿日期:2020-11-03 发布日期:2022-04-27
  • 通讯作者: 杜猛俊 E-mail:du_mengjun@zj.sgcc.com.cn
  • 基金资助:
    国网浙江省电力有限公司科技项目(5211HZ19014U)

UAV-vision-based moving target tracking scheme in electric scenario

FENG Xue, DU Mengjun, XIANG Xinyu, QIAN Jin, ZHANG Min   

  1. Hangzhou Power Supply Company of Zhejiang Power Co. Ltd. of State Grid Corporation of China, Hangzhou 310000, China
  • Received:2020-11-03 Published:2022-04-27

摘要: 随着人工智能技术的发展,面向电力系统的运动目标追踪技术逐渐得到关注,现有方法虽有一定成效,但是大多基于固定摄像头的监控视频录制,不能灵活追踪运动目标,当运动目标离开摄像头视野时,存在运动目标丢失问题。为此,利用无人机设备,并基于深度学习和核相关滤波技术,提出了一个电力场景下基于无人机视觉的运动目标追踪方法(MTTS_UAV)。所提方法采用改进的目标追踪方法与目标检测方法相结合的方式来追踪运动目标隐患,并引入2种无人机飞行控制模块:启发式和数据驱动式,使得无人机的飞行速度和方向可以根据目标移动情况自适应地调节。在真实变电站的安全帽人员数据集上进行了大量实验,对所提方法的追踪效果进行评估,结果表明:所提方法在真实数据集上的平均像素误差(APE)和平均重叠率(AOR)分别可达到2.37和0.67,验证了方法的有效性。

关键词: 目标追踪, 目标检测, 深度学习, 无人机视觉, 电力场景

Abstract: With the rapid development of artificial intelligence technology, the moving target tracking technology for the power system has gradually attracted researchers' attention. Although the existing methods have achieved great success, most of them are based on the fixed camera surveillance video recording, which cannot track the moving target flexibly. When the moving object leaves the camera's field of view, there is a problem of losing the moving object. In light of this, we propose a moving target tracking scheme based on UAV vision (MTTS_UAV) in electric scenario. In particular, to ensure the real-time feature, we combine the improved target tracking algorithm and target detection algorithm to track the hidden dangers. To ensure the accuracy, we introduce two UAV flight control modules:heuristic mode and data-driven mode, so that the UAV's flight speed and direction can be adjusted adaptively according to the target movement. Extensive experiments have been conducted on a real-world dataset of hardhat personnel in real substations, which demonstrate that the average pixel error (APE) and average overlap rate (AOR) are 2.37 and 0.67 respectively and verify the effectiveness of the proposed method.

Key words: target tracking, target detection, deep learning, UAV vision, electric scenario

中图分类号: 


版权所有 © 《北京航空航天大学学报》编辑部
通讯地址:北京市海淀区学院路37号 北京航空航天大学学报编辑部 邮编:100191 E-mail:jbuaa@buaa.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发