A dataset for Wi-Fi-based human-to-human interaction recognition

This paper presents a dataset for Wi-Fi-based human-to-human interaction recognition that comprises twelve different interactions performed by 40 different pairs of subjects in an indoor environment. Each pair of subjects performed ten trials of each of the twelve interactions and the total number of trials recorded in our dataset for all the 40 pairs of subjects is 4800 trials (i.e., 40 pairs of subjects × 12 interactions × 10 trials). The publicly available CSI tool [1] is used to record the Wi-Fi signals transmitted from a commercial off-the-shelf access point, namely the Sagemcom 2704 access point, to a desktop computer that is equipped with an Intel 5300 network interface card. The recorded Wi-Fi signals consist of the Received Signal Strength Indicator (RSSI) values and the Channel State Information (CSI) values. Unlike the publicly available Wi-Fi-based human activity datasets, which mainly have focused on activities performed by a single human, our dataset provides a collection of Wi-Fi signals that are recorded for 40 different pairs of subjects while performing twelve two-person interactions. The presented dataset can be exploited to advance Wi-Fi-based human activity recognition in different aspects, such as the use


a b s t r a c t
This paper presents a dataset for Wi-Fi-based human-tohuman interaction recognition that comprises twelve different interactions performed by 40 different pairs of subjects in an indoor environment. Each pair of subjects performed ten trials of each of the twelve interactions and the total number of trials recorded in our dataset for all the 40 pairs of subjects is 4800 trials (i.e., 40 pairs of subjects × 12 interactions × 10 trials). The publicly available CSI tool [1] is used to record the Wi-Fi signals transmitted from a commercial off-the-shelf access point, namely the Sagemcom 2704 access point, to a desktop computer that is equipped with an Intel 5300 network interface card. The recorded Wi-Fi signals consist of the Received Signal Strength Indicator (RSSI) values and the Channel State Information (CSI) values. Unlike the publicly available Wi-Fi-based human activity datasets, which mainly have focused on activities performed by a single human, our dataset provides a collection of Wi-Fi signals that are recorded for 40 different pairs of subjects while performing twelve two-person interactions. The presented dataset can be exploited to advance Wi-Fi-based human activity recognition in different aspects, such as the use of various machine learning algorithms to recognize different human-to-human interactions.
© 2020 The Author(s Value of the data • The dataset contains a collection of Wi-Fi signals, including the RSSI and CSI values, which are recorded for 40 different pairs of subjects while performing twelve human-to-human interactions in an indoor environment. To the best of our knowledge, this is the first Wi-Fibased dataset that considers the activities performed by two individuals. • Researchers in the field of human activity recognition can utilize the acquired data to evaluate the performance of Wi-Fi-based human-to-human interaction recognition systems that can be developed for various application domains. • The acquired data can be exploited to advance human activity recognition technology in different aspects. For example, various pattern recognition and machine learning methods can be used to accurately recognize different human-to-human interactions. Another potential use of our data is to explore the utilization of various signal processing techniques to analyze the recorded Wi-Fi signals and extract salient features that can be used to recognize different human-to-human interactions. • The existing publicly available Wi-Fi-based human activity datasets have focused on activities that are performed by a single human. On the contrary, our dataset contributes to the ongoing research in the field of Wi-Fi-based human activity recognition by providing a collection of Wi-Fi signals that are recorded from 40 different pairs of subjects while performing twelve human-to-human interactions.

Data description
The raw data are grouped into one main folder that comprises 40 subfolders, where each sub-folder contains the data files recorded for a particular pair of subjects. Specifically, a total of 120 trials (i.e., 10 trials per each interaction x 12 interactions) were recorded for each pair of subjects, where each trial was stored in a separate MATLAB data file (.mat).
The name of each data file follows the form "Sx_Sy_In_Tk.mat". The first part of the name of each data file, denoted as Sx_Sy, represents the pair of subjects who have performed the interaction recorded in a specific data file. In particular, Sx and Sy are the first and second subjects within the pair Sx_Sy, respectively. The pairs of the subjects were formed from a pool of 66 different subjects, as described in the next section. Hence, x and y are integer numbers between 1 and 66. The second part of the name of each data file, denoted as ln, represents one of the twelve human-to-human interactions, where n is an integer number between 1 and 12. Specifically, In = {approaching (I1), departing (I2), handshaking (I3), high five (I4), hugging (I5), kicking with the left leg (I6), kicking with the right leg (I7), pointing with the left hand (I8), pointing with the right hand (I9), punching with the left hand (I10), punching with the right hand (I11), and pushing (I12)}. Finally, the last part of the name of each data file, denoted as Tk, represents the trial number, where k is an integer number between 1 and 10. For example, the data file name "S15_S3_I2_T3.mat" represents a file containing the data recorded for the pair of subjects S15_S3 while performing the departing interaction during the third trial.
The data file associated with each trial contains a cell array of dimension L × 1, where L represents the number of Wi-Fi packets captured during the recording of a particular trial. Moreover, each Wi-Fi packet is stored within an element of the cell array in the form of a structure that consists of several fields as described in Table 1 . Fig. 1 shows the average ± standard deviation number of Wi-Fi packets recorded in all the intervals within a trial, the interaction interval within a trial, and the steady state interval within a trial computed for each of the twelve interactions over all the pairs of subjects. Fig. 2 shows the CSI signals recorded for the pair of subjects that constitutes subject 6 and subject 24 while performing the twelve human-to-human interactions. Moreover, the threedimensional mesh plots presented in Fig. 2 show the different intervals comprised within each of the twelve interactions, including the steady state and human-to-human interaction intervals.

Subjects
A total of 66 healthy subjects (63 males and three females, average ± standard deviation age of 22.1 ± 3.7 years) have volunteered to participate in the experiments. All subjects received a thorough explanation of the experimental procedure. The experimental procedure was conducted according to the Declaration of Helsinki and approved by the research ethics com- Table 1 The description of the fields in the stucture that contains a Wi-Fi packet.

Field Description timestamp_low
The arrival time of the Wi-Fi packet, which is represented by the lower 32 bits of NIC's clock [1] . This timestamp also represents the arrival time of the RSSI and CSI values comprised within the Wi-Fi packet. Nrx Nrx represents the number of antennas used at the receiver side (i.e., the NIC) and its value is set to 3. Ntx Ntx represents the number of antennas used at the transmitter side (i.e., the access point) and its value is set to 2. noise The measured noise over the channel. agc Represents the automatic gain control parameter of the NIC measured in dB. The value of this field along with the value in the noise field are necessary to convert the unit of the RSSI values from dB to dBm as described in the CSI tool [1] .

RSSI_a
RSSI_a represents the RSSI value received at the first antenna of the NIC measured in dB.

RSSI_b
RSSI_b represents the RSSI value received at the second antenna of the NIC measured in dB.

RSSI_c
RSSI_c represents the RSSI value received at the third antenna of the NIC measured in dB.

CSI
The channel state information in the form of a complex three-dimensional matrix that has a dimension of Ntx × Nrx × Nsc. Nsc represents the number of subcarriers constructed using the Orthogonal Frequency-Division Multiplexing (OFDM) modulation scheme, which is applied to the utilized 20 MHz wide channel. The CSI tool specifies the value of the Nsc parameter to 30 subcarriers [1] . label The recorded trial for any of the twelve human-to-human interactions consists of two types of intervals, namely the steady-state and the interaction intervals. During the steady state interval, the pair of subjects are standing against each other without performing any activity. On the other hand, during the interaction interval, the pair of subjects perform one of the twelve different human-to-human interactions. Thus, this field assigns a label to the Wi-Fi packet to specify whether the packet has arrived during the steady state interval or the interaction interval. In particular, the assigned label is a string of the form In, where n is an integer in the range of 1 to 13 that is assigned to each Wi-Fi packet as follows:  mittee at the German Jordanian University. A signed consent form was collected from each subject.
To perform the twelve human-to-human interactions, we have constructed 40 different pairs of subjects from the 66 subjects who have volunteered to participate in this experiment. In particular, the pairs of subjects were constructed according to the following criteria [4] : (1) each subject was selected to be a member of at most two different pairs of subjects, and (2) each subject that was selected as a member of two different pairs has to have different roles in the two pairs, where the role of a subject can be either active role or passive role depending on whether the subject has initiated the interaction or not. Table 2 shows the constructed pairs of subjects along with the role, gender, age, height, and weight of the subjects within each pair.

Experimental procedure
Each pair of subjects was asked to perform ten different trials of each of the twelve humanto-human interactions. Fig. 3 shows sample images of the twelve human-to-human interactions considered our dataset. Each of the twelve human-to-human interactions consists of two types of intervals, namely the steady state and the interaction intervals. During the steady state interval, the pair of subjects are standing against each other without performing any activity. On the other hand, during the interaction interval, the pair of subjects perform one of the twelve different human-to-human interactions.
In order to accurately perform the different human-to-human interactions, we have designed twelve timing diagrams that describe how to perform each of the twelve human-to-human interaction. Moreover, we developed a group of pre-programmed beep sounds, where each one of these sounds can be played at a preset time instance to notify the subject to perform a specific action during the time interval following the beep sound. In particular, a short beep sound is used to initiate the recording of each trial, a medium beep sound is used to indicate an interval transition, and a long beep sound is used to announce the end of the recording of each trial. Table 3 shows the timing diagram associated with each of the twelve human-to-human interactions along with the steady state interval, interaction interval, and the time instances associated with the added beep sounds. These timing diagrams were thoroughly explained to the subjects before the beginning of data recording. Moreover, the subjects were asked to follow the timing diagrams during the performance of the twelve interactions and to perform their roles within the amount of time allocated to each interval of a particular interaction. Table 2 The constructed pairs of subjects along with the role, gender, age, height, and weight of the subjects within each pair.

Table 3
The timing diagram associated with each of the twelve human-to-human interactions along with the steady state interval, interaction interval, and the time instances of the added beep sounds. A sound icon is used to mark the locations of the added beep sounds.   [ 2 , 3 , 5 ].
The access point comprises two internal transmit antennas (i.e., Ntx = 2), and the NIC has three external receive antennas (i.e., Nrx = 3). Therefore, the resultant MIMO system consists of 2 × 3 Wi-Fi streams, where each MIMO stream is established between a unique pair of transmitreceive antennas. Moreover, for each OFDM-modulated MIMO stream, the CSI tool is capable of capturing the CSI for 30 subcarriers (i.e., Nsc = 30) that are evenly spread over the selected channel bandwidth, which is equal to 20 MHz. Thus, our MIMO system is capable of capturing 6 × 30 subcarriers. Fig. 5 shows the MIMO streams established between the utilized access point and the NIC, which are used to record the CSI while the subjects are performing the twelve human-to-human interactions.

Environment
The Wi-Fi signals were captured in a furnished room of dimensions 5.3 m × 5.3 m, as shown in Fig. 6 . The access point and the NIC were mounted in a line-of-sight configuration at a distance of 4.3 m apart from each other. The pairs of subjects performed the twelve humanto-human interactions in the center of the area located between the access point and the NIC.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships which have, or could be perceived to have, influenced the work reported in this article.