WiFi (CSI and RSSI) Data of three Fine-grained Knife Activities (Chop, Slice, French Cut) in an Authentic Kitchen with Different Sensor Placements

Device-free human activity recognition (HAR) has been among the commonly investigated HAR methods in the past few years. Even so, a neglected area in this field is the sensor placement in authentic environments. This paper presents a research method to investigate the effects of sensor placement in the recognition performance of fine-grained human activities in an authentic dining context. For this purpose, we proposed three sensor placements by having user and performing area in line-of-sight (LOS) and non-line-of-sight (NLOS).The dataset contains CSI and RSSI data for 3 activities in 3 sensor placements.It is in a zip file containing 3 folders named placement 1, 2, and 3.Each folder contains 31 files for different activities (preparation, chop, slice, and French cut)Each file contains a list of data entries.Each data entry contains the following parts:type (CSI_DATA) - role (AP)- mac (96:DC:5A:81:E5:46) - rssi (-59) - rate (11) - sig_mode (1) - mcs (4) - bandwidth (0) - smoothing (1) - not_sounding (1) - aggregation (0) - stbc (0) - fec_coding (0) - sgi (0) - noise_floor (-95) - ampdu_cnt (0) - channel (8) - secondary_channel (0) - local_timestamp (1718710) - ant (0) - sig_len (126) - rx_state (0) - real_time_set (0) - real_timestamp (2.45605) - len (256) - CSI_DATA (a list of numbers) CSI_DATA in each data entry contains 128 numbers (64 complex numbers), The middle 11th complex numbers (22 numbers) which are 0 and we called the null subcarrier. In addition, the first complex number (2 numbers) is not valid because they are "LO Leakage". So, we will have 128 – 24 = 104 numbers which are 52 complex numbers. These are our 52 subcarriers of CSI data.