Scholarlink

<p>Parking scarcity in megacities generates disproportionate congestion externalities, driven not only by capacity limitations but by behavioral uncertainty and mixed-traffic occlusion effects that undermine conventional sensor-driven parking management. This study presents the optimal parking space allocation model (OPSAM), an AI-augmented parking intelligence framework that unifies computer vision, behavioral modeling, operational metadata, and spatio-temporal learning for real-time occupancy detection and short-term forecasting across 11 districts of Delhi. The dataset integrates ~ 20,000 manually annotated CCTV frames, 6&#xa0;months of historical occupancy records, detailed parking inventories, and 1550 district-wise behavioral responses. The YOLOv8 visual module demonstrated strong resilience to occlusions and illumination variance (mAP@0.5 = 0.94; F1 = 0.915), with a 95.7% correct IoU-based slot assignment. The CNN–LSTM hybrid prediction model achieved high near-term forecasting accuracy (R<sup>2</sup> = 0.93 at 5-min horizon; RMSE = 0.08), while incorporation of the composite behavior index, derived from cost sensitivity, walking tolerance, search tolerance, duration patterns, and perceived difficulty, reduced forecast error by 14–18%. Random Forest-based cruising-time estimation attained R<sup>2</sup> = 0.87, quantifying behavioral and occupancy-driven search delays. District-wise comparisons revealed distinct parking cultures highest difficulty and illegal-parking bias in Central and West Delhi, versus smoother demand cycles in New Delhi’s institutional corridors emphasizing the inadequacy of uniform regulatory prescriptions. OPSAM provides a scalable foundation for dynamic pricing, targeted enforcement, adaptive occupancy guidance, and demand-responsive curb allocation in dense mixed-traffic environments. While current behavioral inputs are cross-sectional, future longitudinal collection and sensor-complemented night-vision enhancements can further strengthen operational deployment.</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

AI-Augmented Real-Time Parking Occupancy Detection and Prediction: Integrating Computer Vision with User Behavioral insights for Delhi’s mixed-traffic ecosystem

  • 期刊论文

  • 发表日期：

    2026年4月29日

查看全文
  • Abdul Ahad,
  • Farhan Ahmad Kidwai,
  • Saeed Alqadhi

摘要

Parking scarcity in megacities generates disproportionate congestion externalities, driven not only by capacity limitations but by behavioral uncertainty and mixed-traffic occlusion effects that undermine conventional sensor-driven parking management. This study presents the optimal parking space allocation model (OPSAM), an AI-augmented parking intelligence framework that unifies computer vision, behavioral modeling, operational metadata, and spatio-temporal learning for real-time occupancy detection and short-term forecasting across 11 districts of Delhi. The dataset integrates ~ 20,000 manually annotated CCTV frames, 6 months of historical occupancy records, detailed parking inventories, and 1550 district-wise behavioral responses. The YOLOv8 visual module demonstrated strong resilience to occlusions and illumination variance (mAP@0.5 = 0.94; F1 = 0.915), with a 95.7% correct IoU-based slot assignment. The CNN–LSTM hybrid prediction model achieved high near-term forecasting accuracy (R2 = 0.93 at 5-min horizon; RMSE = 0.08), while incorporation of the composite behavior index, derived from cost sensitivity, walking tolerance, search tolerance, duration patterns, and perceived difficulty, reduced forecast error by 14–18%. Random Forest-based cruising-time estimation attained R2 = 0.87, quantifying behavioral and occupancy-driven search delays. District-wise comparisons revealed distinct parking cultures highest difficulty and illegal-parking bias in Central and West Delhi, versus smoother demand cycles in New Delhi’s institutional corridors emphasizing the inadequacy of uniform regulatory prescriptions. OPSAM provides a scalable foundation for dynamic pricing, targeted enforcement, adaptive occupancy guidance, and demand-responsive curb allocation in dense mixed-traffic environments. While current behavioral inputs are cross-sectional, future longitudinal collection and sensor-complemented night-vision enhancements can further strengthen operational deployment.