Journal of Korean Institute of Architectural Sustainable Environment and Building Systems ISSN:1976-6483(Print) 2586-0666(Online) 한국건축친환경설비학회논문집

Occupant-centric control of lighting and HVAC is a key strategy for reducing building energy consumption, and privacy-preserving occupant sensing is a prerequisite for its application to residential spaces. A low-cost, low-resolution 2D LiDAR provides anonymous planar distance measurements, but the number of points returned from a human body decreases with distance, so density-based clustering with fixed parameters cannot maintain consistent detection over an entire room. This study empirically derives distance-adaptive clustering parameters for indoor occupant detection from long-term measurements in an actual office testbed. Based on a theoretical point-count model validated against the measurements, a distance-adaptive minimum- points rule for DBSCAN, decreasing stepwise from 10 to 3 between 0.5 m and 8 m, was derived from the 5th percentile of measured point counts, and cluster-size bounds for rejecting non- human objects were established from the 1st and 99th percentiles of occupant-cluster sizes, as a constant lower bound of 0.148 m and a distance-increasing upper bound of 0.549 + 0.011d m. In a leave-one-day-out cross-validation against tracking-verified occupant positions, the proposed parameters improved recall from 0.904 to 0.967 and precision from 0.973 to 0.980 relative to conventional fixed parameters, providing a calibrated detection front-end for occupant trajectory tracking and occupant-centric lighting and ventilation control.