Data and Sourcecode from: Neural Network-based Occupancy Detection on the Edge

Environmental Data Collected for Data-Driven Occupancy Detection

Dataset Information

The following data is collected from LoRa sensors of two rooms for a period of three months in an office building on the ground floor in Graz, Austria:

Open status of windows/doors

Relative humidity

CO2 concentration

Ambient temperature

PIR-based motion counter

Light level

IR-based occupancy (only room A)

Average/peak sound level

Radar-based people counter (left-to-right and right-to-left; only room A; no trustworthy ground truth!)

Folder Organization in occupancy-detection-dataset.zip

    ├── data

    │   ├── interim       <- Intermediate data of room A and B that has been transformed.

    │   └── raw            <- The original, immutable sensor data dump of room A and B.

Raw Data

Raw sensor data of room A and B consisting of six and two work places respectively. Data is gathered in an interval of five minutes.

Note:

Timezone ist UTC+00:00.

Column "occupancy" in df_features.csv refers to IR based occupancy sensor from Elsys ERS Eye (Possible values 0-2).

Column "motion" in df_features.csv refers to a PIR based motion counter.

IR-based occupancy is not measured in room B.

Intermediate Data

Event-based (door and window sensors) and interval based (humidity, CO2, temperature, ....) data is synchronized to retrieve a homogenous data set.

Window columns are merged to represent the number of open windows. Nothing else was applied to the data.

Ground Truth

Image-based occupancy ground truth data is separated in a file (df_occ.csv).

It describes the number of occupants at a certain time stamp provided from images (manually labelled).

References

Coming soon.