Day-in-the-Life: a full-day, pose-resolved indoor pedestrian trajectory dataset

A single complete local calendar day of pedestrian tracking from one indoor retail environment, released in full rather than as a curated excerpt. The day runs 00:00:00 to 23:59:59 local time and includes the overnight period when the site is unoccupied. 2,109,812 observations 1,817 distinct tracks 288 of 288 five-minute chunks present, no gaps; 49 chunks contain no observations (genuine overnight vacancy, between 00:50 and 06:45 local) Every observation carries a 24-joint 3D body pose in addition to floor-plane position, velocity and acceleration Two representations, not equivalent: canonical_tracks.parquet holds 22 kinematic columns and no pose; chunks_csv.zip expands to 288 chunk_local_*.csv files holding those same 22 columns plus 72 pose columns. Both contain the same 2,109,812 rows and 1,817 tracks. Taking only the parquet discards all pose data. The chunk CSVs are bundled into a single zip because Zenodo permits at most 100 files per record; unzip in place to recover the individual files. MANIFEST.sha256 covers this record as deposited and MANIFEST_EXPANDED.sha256 covers all 293 files after extraction. A trajectory is one person_track_id within this day; the release establishes no cross-day identity and track ids are not person identifiers. The calendar date is a surrogate: time-of-day and intra-day sequence are exact, but the date carries no information. Data are released de-identified; re-identification is prohibited by the license. See README.md for the full schema, identity scope, and analysis guidance (notably: zero-fill the empty intervals when binning by time).