Smart Camera Parking System With Auto Parking Spot Detection

Tuan T. Nguyen, Mina Sartipi; Proceedings of the Asian Conference on Computer Vision (ACCV) Workshops, 2024, pp. 232-241

The proliferation of urban centers has exacerbated traffic congestion, underscoring the critical need for intelligent parking solutions. While computer vision approaches have gained traction, their reliance on manual spot labeling poses practical challenges. This study introduces PakLoc, a novel framework for automated parking spot detection, complemented by PakSke, a module that refines bounding box orientation and dimensions. Empirical evaluation on the PKLot dataset demonstrates a remarkable 94.25% reduction in manual labor requirements. Furthermore, we present PakSta, an innovative method leveraging PakLoc's object detector to concurrently assess occupancy across all parking spaces within a given frame. PakSta achieves an impressive AP75 of 93.6% on the PKLot dataset, surpassing the performance of the benchmark Yolo SPS (93.3% AP75) which relies on manually labeled data, and significantly outperforming other methods such as POD (61.8% AP75). These advancements offer a promising avenue for efficient, label-free smart parking systems, potentially revolutionizing urban parking management.