From Rooms to Rows: Adapting Indoor Functional-Zone Segmentation to Open Parking Environments

Indoor 3D scene-graph methods increasingly subdivide open-plan environments into functional sub-regions by projecting scanned geometry into a bird’s-eye-view representation and applying distance-field and watershed-based segmentation. This work investigates whether the underlying functional-zone concept can be transferred to a structurally similar but domain-different setting: open, wall-free parking environments, where meaningful spatial organization emerges from contiguous clusters of occupied and free parking spaces rather than from individual spaces alone. We formalize the indoor Euclidean distance field (EDF)/watershed formulation and introduce a lightweight connectivity-based analogue designed for discrete, pre-annotated occupancy data. The proposed formulation is adapted to outdoor parking environments using the PKLot dataset, with occupancy annotations converted into spatial zone maps and grouped through morphological dilation and connected-component analysis. A full-day sequence from a single fixed camera is then analyzed temporally using both functional-zone counts and a proposed fragmentation index that quantifies the spatial organization of occupancy beyond raw occupied/free-space counts. The resulting zone counts broadly follow the temporal trends of raw parking occupancy while revealing additional spatial structure during periods of fragmented, checkerboard-like occupancy. In particular, two frames with different spatial configurations exhibit a 2.50× increase in the free-space fragmentation index (0.145 to 0.362), demonstrating that similar occupancy statistics can correspond to substantially different spatial organizations. The results suggest that functional-zone representations can provide a lightweight descriptor of parking-environment structure that is not captured by conventional occupancy counts alone. This study is intentionally scoped as a first-pass adaptation rather than a complete perception or safety system. We identify three key extensions required for deployment in vehicle-yard and loading-dock scenarios: learned CNN-based occupancy inference from raw imagery, full BEV distance-field/watershed boundary estimation, and paired vehicle-trajectory data for evaluating zone transitions during backing and maneuvering. The work therefore provides a preliminary bridge between indoor functional-zone segmentation and outdoor, vehicle-relevant spatial reasoning.