Sensor based dataset for the identification of users with a door handle

Abstract

Abstract: This data set consists of series of vectors formed from angular speeds and accelerations, conveniently labeled with time in µseconds. A total of 47 individuals (13 female and 34 male) aged from 18 to 68 years old collaborated in the making of our database. Independently of whether the subject was right- or left-handed, the individuals were asked to interact as naturally as possible with a right-handed door handle for 25 repetitions on a firmly closed door. The subjects were asked to try to open a closed door: (i) to start at a distance of 3 m right in front of the door and walk as they usually will do with the intention of acting on the door handle, then (ii) to act on the door handle, (iii) to wait a very short time with the lever on the end-stop position, (iv) to release the force while handling the door handle, and finally (v) to move away the hand. For each of the donors, age, gender and height were recorded and stored aside the samples. No other personal information was included on it. All the persons were informed about the purpose of the corpus, and that the recorded data would be made available for the community in an anonymous way, and they signed an uninterested agreement donating of the data. The first 200 samples of each series capture the rest state of the system before the instant in which the platform detects a significant change in the energy of the signal; from this very moment the platform acquires a fixed amount of samples up to complete a total duration of 2.5 s, that we considered large enough to gather all data. Discarding some series that did not contain the whole user interaction with the door handle was a minor problem, and finally a corpus consisting of 20 repetitions for each subject was created and labeled, resulting of a total of 960 attempts of opening a door.