Prof. Panos Nasiopoulos
The University of British Columbia, Canada
Title of the paper: Navigating Urban Gridlock: An Innovative Real-Time Street Parking Solution for Human- Driven and Autonomous Vehicles
Abstract: A novel real-time street parking solution is presented to alleviate urban traffic congestion caused
by drivers searching for parking spaces—an activity estimated to account for up to 50% of traffic
in metropolitan areas.
The proposed holistic system integrates on-vehicle cameras, embedded parking and sign
recognition models, and edge computing to automatically detect, localize, and communicate the
availability of free parking spots to nearby vehicles within a 3–4 km radius.
A critical component of the system is the precise labeling and measurement of parking spaces,
achieved through motion vector analysis and geometric calibration. Experimental evaluation
demonstrated 91.4% accuracy in parking space measurement and an overall 87.5% detection
accuracy for identifying available street parking.
The system further incorporates parking sign detection and identification using a two-stage
process:
1. Object detection via a lightweight YOLOv7-X model, selected for its optimal balance
between inference speed and accuracy; and
2. Sign recognition and classification using a matching network trained with triplet loss.
A reference database containing 43 classes and 116 sample images of parking signs was
developed to improve recognition robustness under diverse viewing angles and lighting
conditions. Subsequent integration of text detection increased the overall sign recognition
accuracy from 81% to 91% through image enhancement and multi-frame verification.
This end-to-end framework demonstrates the feasibility of deploying intelligent, camera-based
parking management systems for both human-driven and autonomous vehicles, offering a
scalable solution to reduce traffic congestion, enhance urban mobility, and support nextgeneration
smart city infrastructures.
Bio: To be announced soon