• ISSN: 1674-7461
  • CN: 11-5823/TU
  • 主管:中国科学技术协会
  • 主办:中国图学学会
  • 承办:中国建筑科学研究院有限公司

基于YOLOv3的无人机建筑物空间特征提取方法研究

Research on Extraction Method of UAV Building Spatial Features Based on YOLOv3

  • 摘要: 随着智慧城市的发展,对建筑空间特征的采集速度和精度要求越来越高,针对市面上常见的空间特征提取方法灵活性差、智能化程度低、人力成本高等问题,本文提出了一种基于无人机扫描与YOLO识别的检测方法,完成了建筑物窗户构件的实时识别与提取。本文对YOLOv3算法进行了优化调整并运用自制的数据集对算法模型进行训练,使用Nginx搭建RTMP(Real Time Messaging Protocol)推流服务器接收无人机扫描信息,在地面平台显示实时识别结果,该方法大大提高了识别效率,降低了推流传输延迟,在无人机检测实验中体现出了实时、高效、智能的特点,借助无人机的灵活性,能够高效地满足建筑构件实时检测的需要。

     

    Abstract: With the development of smart cities, the requirement of sampling spatial characteristics of building regarding to efficiencies and accuracies are becoming much higher.Focusing on those limitations of current available technologies in AEC market, which is poor of flexibilities, lack of intelligences, and high of labor cost, this paper has proposed a new solution based on UAV(Unmanned Aerial Vehicle)scanning and YOLO(You Only Look Once)recognition to enable the real-time identification and extraction of window components from as-built projects. In this paper, the algorithm of YOLOv3 is being optimized to train the algorithm model by using integrated dataset. Nginx is being used to build RTMP(Real Time Messaging Protocol)for receiving the data from UAV scanning through streaming server for displaying the live results in ground station. The proposed solution has improved the efficiencies of recognition and has reduced the delays of streaming, which reflects the advantages of UAV technologies from the aspects of simultaneous, efficiencies, and intelligences for satisfying those demands in building components detection.

     

/

返回文章
返回