Abstract Manager

Urban curbside parking availability is constrained not only by parked vehicles and regulatory signage but also by physical access infrastructure, such as driveways and alleys, that reduce demand and impose mandatory clearance requirements. In residential streets, vehicles are prohibited from parking within a defined distance of these access points, making their identification and spatial extent essential for realistic estimates of parking utilization. However, large-scale mapping of such infrastructure is rarely incorporated into parking analytics due to the lack of scalable data collection and processing methods.

This study develops and evaluates a mobile vision–GPS framework for segmenting and spatially mapping curbside infrastructure elements that directly influence parking availability, with a focus on driveways and alleys in the City of Edmonton’s Residential Parking Program. High-resolution 4K video data are collected using a vehicle-mounted GoPro camera, capturing curbside environments under real-world operating conditions. Driveways are classified by functional type (single, double, and triple access), enabling differentiation of their spatial footprint and regulatory impact on curbside parking eligibility.

Infrastructure segmentation is performed using a deep learning–based semantic segmentation framework built upon the Mask2Former architecture, initialized with pre-trained weights. A comprehensive experimental design evaluates the sensitivity of model performance to training duration, learning rate magnitude, and decay strategies, optimizer selection, gradient clipping, validation frequency, and class oversampling strategies. Multiple training configurations are tested to assess robustness and generalization across diverse residential contexts.

Model performance is evaluated using standard segmentation metrics, with particular attention to class-level performance for smaller or less frequent infrastructure elements. The resulting segmented infrastructure layers will be spatially aligned with GPS trajectories, enabling the quantification of curbside segments affected by driveway and alley constraints and the estimation of effective parking availability after regulatory clearances are applied.

The results aim to demonstrate that explicitly accounting for curbside access infrastructure leads to more realistic estimates of parking utilization and available curbside space. By integrating infrastructure segmentation with vehicle- and signage-based analytics, the proposed framework provides municipalities with a scalable, data-driven tool to support residential parking design, regulation, and enforcement.