Parking availability remains a persistent challenge in urban and institutional environments, particularly on university campuses with limited space and high demand. This work presents a vision-based detection system using YOLOv12 object detection models to identify vacant parking spots. Two model variants, YOLOv12-Medium (YOLOv12m) and YOLOv12-Large (YOLOv12l), were trained on a class-balanced dataset of 18,698 images, including diverse lighting conditions. A comparative evaluation revealed that YOLOv12m achieved a lower average detection time per image (4.86 ms) and a higher F1–score (0.984), whereas YOLOv12l provided marginal robustness under low-light scenarios. Based on this analysis, YOLOv12m was integrated into a modular web-based interface for continuous monitoring of parking occupancy. Campus deployment confirmed the system’s usability, responsiveness, and potential for a scalable application. The proposed solution addresses key requirements for intelligent parking systems, including generalization, ease of deployment, and user-friendly interaction.