Deep Learning-Based Real-Time Parking and Driver Disembarkation Detection System in No-Parking Zones

Traditional illegal parking enforcement operates on a post-detection basis, meaning that violations are identified long after the driver has left the vehicle. This approach has limitations in ensuring smooth traffic flow and pedestrian safety. To address these issues, this paper proposes a system that integrates YOLO-based object detection, the Darknet framework, and MOSSE-based Multi-Object Tracking (MOT) technology. The system identifies vehicle parking events and detects driver exit events from surveillance footage. The proposed approach achieves over 91% detection accuracy for identifying driver exit moments. Instead of relying on license plate-based enforcement, this system aims to provide proactive guidance by immediately issuing on-site warnings and notifications before the driver moves away from the vehicle, encouraging prompt vehicle relocation and reducing illegal parking incidents.