A High-Fidelity Residential Building Occupancy Detection Dataset

This repository contains data that was collected by the University of Colorado Boulder, with help from Iowa State University, for use in residential occupancy detection algorithm development. Data was collected with the mobile human presence detection system (HPDmobile) for residential occupancy detection. Over the course of the project, data, which included images, sound, and indoor environmental readings, was collected from inside of six residential homes, from a minimum of four locations concurrently in each home. Audio and images have been processed to remove personally identifiable information. Data was only collected from common areas (kitchens, offices, living rooms). The data-set contains 15-35 days of data from each of the six homes, and consists of the following readings, taken from a minimum of four different locations inside each home: - 32-by-32 pixel grayscale png images, captured every second - 10-second long audio files which were processed to remove identifiable speech, and converted to text - Indoor environmental reading, which include temperature, relative humidity, illuminance, equivalent CO2concentration, and total volatile organic compounds, all captured every ten seconds - Validated, ground truth binary occupancy status of the home - Lists of images that were captured, but not included, as the image was very dark (as determined by average pixel value) - Zone labels for each image, specifying if a human is in the image or not (binary) - Metadata (csv files) with statistics on completeness of the data collected, hardware used, homes monitored, and accuracy of the zone based images, per home and per hub, along with a txt file describing the metadata With the exception of images, all data is stored as csv files within compressed files. All data are separated by modality type. Compressed image files contain all data for one hub, while the other modality zip files contain data for all hubs.