Automated Recognition and Localization of Parking Signs Using Street-Level Imagery

Parking signs recognition and localization aim to extract and digitize accurate on-street parking restrictions. Today’s visual data collection, annotation, and analysis practices are still costly, prone to error, and cumbersome as performed manually. While online street-level imagery databases contain updated panoramic images of all signs, their potential for understanding on-street parking restrictions at scale has not been fully explored. The key benefit of these databases is that once the parking signs are detected, accurate geographic coordinates of the detected signs can be automatically determined and visualized within the same platform. This paper evaluates the application of a computer vision-based method for parking signs recognition from street-level imagery aimed at facilitating the parking in dense cities. The method extracts images and leverages a sliding window mechanism to detect potential candidates for parking signs. A Histogram of Oriented Gradients for each candidate is formed and trained. Hard negative mining approach is then used at final stage to remove the remaining false positives. The potential detections of images from multiple viewpoints are then combined to locate the position of the signs on a map. Experimental results with high accuracy demonstrate the potential of leveraging street-level images and provide a viable solution for digitizing at scale all parking signs to help drivers understand parking rules and avoid fines.

Language

  • English

Media Info

  • Media Type: Web
  • Features: Figures; References; Tables;
  • Pagination: 9p
  • Monograph Title: Computing in Civil Engineering 2017: Information Modeling and Data Analytics

Subject/Index Terms

Filing Info

  • Accession Number: 01683657
  • Record Type: Publication
  • ISBN: 9780784480823
  • Files: TRIS, ASCE
  • Created Date: Oct 4 2018 4:26PM