A Data Fusion Approach for Extracting Highway Maintenance Features

Most existing studies investigating highway feature extraction in the literature focus on a single source of data, e.g., aerial LiDAR or imagery data. Because each source of input data contains its own characteristic visual features, the fusion of the characteristic and complementary features acquired from both sensors can provide more reliable information to facilitate the accurate detection of maintenance features and therefore improve the robustness of the algorithm, especially when data from either of the input sources are of low quality. In the first stage of this project, the authors propose an effective method to automatically detect traffic signs and light poles from mobile LiDAR data in the I-15 highway without any preprocessing or learning steps. A set of extensive experiments have been carried out on the data sets, which are captured by UDOT along the I-15 highway. The results demonstrate the robustness of the proposed method in detecting almost all traffic signs and light poles. In the second stage of this project, the authors propose a novel deep learning method to accurately detect traffic signs by fusing the characteristic and complementary features in automatically registered airborne geo-referenced images and airborne LiDAR data. A set of extensive experiments has been carried out on the airborne geo-referenced images and airborne LiDAR data captured by USU along the I-15 highway. The results demonstrate the robustness of the proposed method in detecting traffic signs.

Language

  • English

Media Info

  • Media Type: Digital/other
  • Edition: Final Report
  • Features: Figures; Photos; References; Tables;
  • Pagination: 49p

Subject/Index Terms

Filing Info

  • Accession Number: 01710881
  • Record Type: Publication
  • Report/Paper Numbers: UT-19.12
  • Contract Numbers: 18-8440
  • Files: NTL, TRIS, ATRI, USDOT, STATEDOT
  • Created Date: Jul 12 2019 12:54PM