Identification of occupant presence and location inside buildings is essential to functional goals such as security, healthcare, and energy management. Floor-vibration measurements, induced by footstep impacts, provide a non-intrusive sensing method for occupant identification, unlike cameras and smartphones. Detecting the presence of an occupant is a necessary first step for occupant location identification. A challenge for occupant detection is ambient noise that may hide footstep-induced floor-vibration signatures. Also, spurious events such as door closing, chair dragging and falling objects may result in vibrations that have similarities with footstep impact events. In this paper, an accurate occupant-detection strategy for structures with varying rigidity is outlined. Event detection is based on computing the standard deviation of a moving window over measurements at various frequency ranges. Using a classification method, footsteps are distinguished from other events. This strategy enhances detection of footstep-impact events compared with methods that employ only thresholds, thereby reducing false positives (incorrect detection) and false negatives (undetected events). Footstep-impact events may then be used for footstep impact localization using model-based approaches. Finally, the utility of this strategy for footstep-event detection is evaluated using a full-scale case study.