SME Building Occupancy Data with Thermal Images and Sensor Readings

Multimode data collected each minute over a period of 3 months within an SME building. The dataset comprises thermal IR images and a text file with normalized CO2, temperature, and motion sensor data from 3 different rooms, all labelled for occupant presence. It was authored by Ruth Ande, last updated on 2026-05-28, and is split into 60% train and 40% test subsets.

Use Cases

• Train occupancy detection models based on labeled thermal IR images and sensor data.
• Develop multimodal fusion algorithms based on synchronized image, CO2, temperature, and motion data streams.
• Validate building energy simulation models based on real-world, minute-level indoor environmental data.
• Analyze patterns of human presence and activity based on time-series sensor data collected over 3 months.

Strengths

• Data is explicitly labeled for occupant presence (positive/negative), providing clear ground truth.
• Includes synchronized multimodal data (thermal images, CO2, temperature, motion) collected each minute.
• Data is pre-split into 60% train and 40% test subsets, facilitating machine learning workflows.
• Covers a continuous 3-month period, offering longitudinal insight into building occupancy.

Limitations

• Row count is unknown, which may limit suitability assessment for large-scale model training.
• Column-level documentation is absent; field semantics must be inferred after download.
• Missing data samples (e.g., CO2) led to the removal of corresponding multimodal data for that time period, potentially reducing dataset size.

Provenance

Source

figshare, author Ruth Ande.

Collection Method

Data collected via sensors and thermal imaging within an SME building.

Time Range

Covers a period of 3 months.

Freshness

Last updated 2026-05-28 16:59:04; freshness should be verified.

Geography

Data collected within an SME building; specific location is unknown.