OPAL: A multimodal ambient sensor dataset for activity, occupancy, and energy analysis in an office workspace

OPAL (Office Presence, Activity and Load) is a multimodal ambient sensor time-series dataset collected over 33 continuous days in a shared university laboratory workspace at the Gwangju Institute of Science and Technology. Two connected rooms were instrumented: Room 207, the main workspace, and Room 205, a meeting and common area. The release combines appliance-level electrical measurements from 21 smart plugs, motion and door-contact events, ambient temperature, camera-generated event states, participant-level activity annotations for three primary participants across 11 workplace activity classes, and room-level occupancy counts that also include non-target laboratory members. The release covers 2026-05-14 14:00 to 2026-06-16 15:00 (KST) and comprises 1,360 daily CSV files holding 33,803,409 records, 1.96 GB in total, across 34 date directories. It ships with a data dictionary, a labelling guideline, quality-control documentation, machine-readable metadata in schema.org, DCAT-AP 2.1 and MLCommons Croissant form, and SHA-256 checksums for every file. Raw video, raw audio, speech content, conversation transcripts, identifiable facial images and participant identity mapping files are not part of the release. Cameras were used for annotation verification only; the only sound-related signal published is the binary sound-event state reported by the Room 205 camera firmware. The dataset supports human activity recognition, occupancy estimation, multimodal sensor fusion, appliance-level energy analysis and privacy-preserving ambient sensing in shared workplaces.