Published January 19, 2026 | Version 1.0

Indoor occupancy ground truth dataset with variable ventilation

  • 1. ROR icon Aalto University
  • 2. ROR icon Helsinki Metropolia University of Applied Sciences

Description

Indoor occupancy ground truth dataset with variable ventilation.

Overview

This dataset contains measurements from a controlled laboratory experiment studying indoor air quality (IAQ) and occupancy detection in a simulated Finnish office environment. The dataset contains several IAQ sensors, and the ventilation rate is also measured. Ventilation rate varies in the dataset.

The data was collected to test ventilation control based on occupant count estimates from an ML model.

Experimental Setup

Test Environment

  • Location: Laboratory designed to simulate a typical Finnish office room
  • Capacity: Up to 4 occupants
  • Dimensions: 2.69m (height) × 2.71m (width) × 3.0m (length)

Ventilation System

  • Type: Variable Air Volume (VAV) system
  • Control: Modulated airflow rates via Air Handling Unit (AHU)
  • Configuration: Supply and exhaust setpoint flows maintained at equal rates
  • Temperature control: Temperature maintained by chiller

Sensors and Measurements

Sensor Measurement Accuracy
Miran DLS IAQ.THB-H9+CO2 CO₂ (ppm) ±30 ppm
  Pressure (mbar) ±1.0 mbar
  Temperature (°C) ±0.3°C
  Relative Humidity (%) ±2.5%
HDK - Duct CO₂ transmitter / controller CO₂ (ppm) ±40 ppm +3 %
Lindab FTCU Airflow (l/s) ±5%
Steinel HPD2 Occupant count  
  • ** Miran DLS IAQ.THB-H9+CO2 placed inside the room**: 1.75 meters
  • ** HDK - Duct CO₂ transmitter / controller: Measures supply and exhaust air CO₂ inside the duct.
  • Occupancy ground truth: Camera-based computer vision occupant counting sensors, and later manually recorded.

Dataset Details

The training data includes multiple ventilation control strategies, and in the test data, we controlled ventilation according to occupant count. 

Occupancy-based control:

  • Base ventilation: 3 l/s (unoccupied)
  • Per-person increment: 6 l/s per detected occupant
  • Update frequency: Every 5 minutes

Training Data

Note that the test room was renovated after March 2025, including the addition of thermal insulation to the walls.

  • Training data 1: February 2024 - March 2025

    • Duration: 5 days, 21 hours
    • The collection saw various usage, meaning that the data may be noisier than the training data 2.
    • Sensors: Miran DLS IAQ.THB-H9+CO2, 2x Lindab FTCU, Steinel HPD2
    • Occupancy data: Camera-based occupancy ground truth
  • Training data 2: October 2025 - November 2025

    • Duration: 2 days, 20 hours days
    • Sensors: Miran DLS IAQ.THB-H9+CO2, 2x Lindab FTCU, 2x HDK - Duct CO₂ transmitter / controller
    • Occupancy data: Manually recorded occupancy ground truth

Ventilation test

  • Period 2: October 2025 - November 2025
    • Duration: 24 hours
    • Collection method: Manually recorded occupancy ground truth

Testing Data

  • Ventilation control testing: 23 hours, 50 minutes

Funding

This research was funded by the Promotion Centre for Electrical Engineering and Energy Efficiency (STEK ry) grant KITSI2.0.

Files

training_data_1.csv

Files (1.4 MB)

Name Size
md5:5a370c371c92d9a186dacd5cf3a0e0d2
817.5 kB Preview Download
md5:c6e22d6b03d224c34b3d3882f0e93b6e
435.3 kB Preview Download
md5:6bd29e069ccce6703cb4df3974741906
152.1 kB Preview Download