Scholarlink

<p>Smart detection of vacant curbside parking can reduce cruising time and congestion, especially in infrastructure-limited urban contexts. This study proposes a Geospatial Vision-Based Deep Learning (GeoVDL) framework that combines low-cost street-level imagery with geospatial reasoning to detect and validate on-street parking availability. In the first stage, a YOLOv8-based object detector identifies vacant spaces in georeferenced camera views, and candidate vacancies are mapped as camera-coverage availability layers in a geospatial engine. In the second stage, a Regulation Context-Aware Geospatial Validation (RCAGV) engine filters visually vacant candidates using regulatory and spatial constraints (e.g., pedestrian crossings, access ramps, restricted curb segments), producing legally admissible parking availability. A case study along a major corridor in Rasht demonstrates reliable performance under diverse urban conditions. The vision module achieved mAP@0.5 of 82.13%, mAP@0.5:0.95 of 35%, and an F1-score of 77.3%. Overall, GeoVDL provides regulation-compliant, spatially actionable curbside parking information for intelligent urban parking management.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Geospatial Vision-Based Deep Learning (GeoVDL) approach for curbside parking detection

  • 期刊论文

  • 发表日期：

    2026年5月12日

查看全文
  • Amirhossein Pardameh,
  • Meysam Effati

摘要

Smart detection of vacant curbside parking can reduce cruising time and congestion, especially in infrastructure-limited urban contexts. This study proposes a Geospatial Vision-Based Deep Learning (GeoVDL) framework that combines low-cost street-level imagery with geospatial reasoning to detect and validate on-street parking availability. In the first stage, a YOLOv8-based object detector identifies vacant spaces in georeferenced camera views, and candidate vacancies are mapped as camera-coverage availability layers in a geospatial engine. In the second stage, a Regulation Context-Aware Geospatial Validation (RCAGV) engine filters visually vacant candidates using regulatory and spatial constraints (e.g., pedestrian crossings, access ramps, restricted curb segments), producing legally admissible parking availability. A case study along a major corridor in Rasht demonstrates reliable performance under diverse urban conditions. The vision module achieved mAP@0.5 of 82.13%, mAP@0.5:0.95 of 35%, and an F1-score of 77.3%. Overall, GeoVDL provides regulation-compliant, spatially actionable curbside parking information for intelligent urban parking management.