Device-free human micro-activity recognition method using WiFi signals

Figures & data

Figure 1. Workflow of the proposed method.

Figure 2. Raw CSI with noise.

Figure 3. CSI after applying lowpass Butterworth filter.

Figure 4. CSI after PCA.

Figure 5. Activity duration width via envelope extraction.

Figure 6. RF flow chart.

Table 1. Experimental details.

Factor	Value
AP	One TP-link WR845N with two antennas
DP	One laptop with IW5300 NIC with three antennas
Collected samples	500 sample for each motion
Sample rates	100 sample/ms for each measurements.
Action position	Fixed position
Types of motions	C1: push, C2: dodge, C3: strike, C4: pull, C5: drag, C6: kick, C7: circle, C8: punch x2(twice), and C9: bowl
Volunteer users	10
Cross validation	10th cross validation

Figure 7. Experimental setup.

Figure 8. Confusion matrix of proposed micro-activities.

Table 2. Precision, recall, and F-measure results.

Class	Precision (%)	Recall (%)	F-measure
C1	94	90.385	92.15
C2	93	90.291	91.625
C3	92	96.842	94.358
C4	90	95.745	92.78
C5	95	92.233	93.59
C6	97	97.98	97.48
C7	98	96.078	97.029
C8	91	88.35	89.655
C9	97	100	98.477

Figure 9. Confusion matrix of proposed micro-activities in NLOS scenario.

Table 3. Precision, recall, and F-measure of each activity in NLOS scenario.

Class	Precision (%)	Recall (%)	F-measure (%)
C1	90	90.9	90.4
C2	88	88.88	88.4
C3	89	89.89	89.4
C4	91	90	90.4
C5	90	87.3	88.6
C6	89	88	88.4
C7	93	93	93
C8	86	85	85.4
C9	89	89	89