[1]王 蓉,吕祖盛,孙 嘉,等.基于人像分割的智能搜救无人机系统设计[J].计算机技术与发展,2020,30(08):147-151.[doi:10. 3969 / j. issn. 1673-629X. 2020. 08. 025]
　WANG Rong,LYU Zu-sheng,SUN Jia,et al.Design of Intelligent Rescue UAV System Based on Portrait Segmentation[J].,2020,30(08):147-151.[doi:10. 3969 / j. issn. 1673-629X. 2020. 08. 025]
点击复制

基于人像分割的智能搜救无人机系统设计()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
30
期数:
2020年08期
页码:
147-151
栏目:
应用开发研究
出版日期:
2020-08-10

文章信息/Info

Title:
Design of Intelligent Rescue UAV System Based on Portrait Segmentation
文章编号:
1673-629X(2020)08-0147-05
作者:
王 蓉; 吕祖盛; 孙 嘉; 江子岍; 肖 建
南京邮电大学 电子与光学工程学院,江苏 南京 210023
Author(s):
WANG Rong; LYU Zu-sheng; SUN Jia; JIANG Zi-qian; XIAO Jian
School of Electronic and Optical Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210023,China
关键词:
无人机救援; 人像分割; 图像拼接; 特征提取; 人体检测
Keywords:
UAV rescue; portrait segmentation; image stitching; feature extraction; human detection
分类号:
TP391
DOI:
10. 3969 / j. issn. 1673-629X. 2020. 08. 025
摘要:
为了在确保救援人员安全的同时,迅速有效地开展野外搜救任务,对搜救方式的选择变得非常重要。 针对传统搜救系统消耗人力大、搜索效率低等弊端,提出了一种无人机智能搜救系统。系统分为自动控制、图像拼接和人像检测三个子系统。控制系统保证无人机依照指定路线安全飞行,实时地采集并回传图像;图像拼接系统将回传的图像通过 ORB特征提取算法合成大型高清航拍影像,以便救援人员能在最短的时间内根据遇难者所在地的地形和环境特点规划最优的救援路线。人像检测系统使用谷歌最新的语义分割模型 DeepLab V3+神经网络进行人像分割,以实现人像检测的功能。 该神经网络运行速度快、精度高、对截断人像识别效果较好,适用于野外救援等紧急情况的实时应用场景。 经过多次实验,结果表明该无人机搜救系统测试稳定,识别准确率高,能在多种复杂的野外环境中发挥较好的救援作用。
Abstract:
In order to quickly and efficiently carry out field search and rescue missions while ensuring the safety of rescue workers,the choice of search and rescue methods becomes extremely important. Aiming at the drawbacks of the traditional search and rescue system, such as high manpower co-nsumption and low search efficiency,an intelligent search and rescue system for UAV is proposed. The system is divided into three subsystems: automatic control,image stitching and portrait detection. The control system ensures that the UAV can fly safely according to the planned route and collect and return images in real time. The image stitching system combines the returned images into large high - definition aerial images by ORB feature extraction algorithm, so that rescuers can plan the best rescue route according to the terrain and environmental characteristics of the loca-tion of the victims in the shortest time. The portrait detection system uses Google’s latest semantic segmentation model DeepLab V3+ neural network for portrait segmentation to realize the function of portrait detection. The neural network has the advantages of fast running speed, high precision and excellent recognition effect for truncated portrait,which is suitable for the real-time application scenarios of field rescue. After many experiments,the results show that the UAV search and rescue system is stable in testing,high in recognition accuracy,and can play a better role in a variety of complex field environment.
更新日期/Last Update: 2020-08-10