An apartment was instrumented with 2 web-cameras in the livingroom and one in the kitchen, plus 3 T-shape microphone arrays per room, each consisting of 4 omni-directional microphones, for a total of 24 audio sensors inside the apartment.
The focus of the data collection was on daily activities that people perform when at home. These activities were grouped into three main categories:
basic activities: phone answering, cleaning-dusting, TV watching, ironing, reading, eating-drinking;
noisy activities, that is single activities as in (1) performed by the subject, X, in the presence of background noise due to the activity of a cognate, Y: X reading while Y is watching TV; X is eating-drinking while Y is watching TV; and X is watching TV while Y is engaged in a phone call;
parallel activities, i.e., two single activities performed simultaneously by the subject: cleaning-dusting + phone answering; ironing + TV watching; eating-drinking + TV watching.
The subjects involved in the data collection were free to choose the location where to perform each activity (the kitchen or the living room) and aspects such as posture (standing, sitting), whether performing it while walking, etc. They had limitations on time, though: each activity was required not to exceed 60 or 90 seconds. The subjects received a message through a headphone, from the experimenters alerting them when it was about time for them to close the current activity, and what the next activity to perform was. They marked the beginning and the end of each activity by touching a PDA screen (in this way an automatic annotation of the collected activities has been accomplished).
20 subjects (13 male and 7 female) participated in the study. Each subject performed the same activity 4 times; hence, the database contains 200 examples per each activity, for a total of more than one hour of recorded data per subject. The total audio-video recordings are about 23 hours and half.
The audio and video recordings, as well as the annotations, are freely available for the research community.
