A Cross-View Gait Recognition Method Using Two-Way Similarity Learning

Table 5

Average recognition rate (%) comparison of the proposed method with DeepCNNs and PoseGait.

Gallery set	DeepCNNs [5]			PoseGait [36]			Ours
	NM #01-04
Probe set	NM#05–06	BG#01–02	CL#01–02	NM#05–06	BG#01–02	CL#01–02	NM#05–06	BG#01–02	CL#01–02
0°	88.7	64.2	37.7	55.3	35.3	24.3	55.7	37.4	26.8
18°	95.1	80.6	57.2	69.6	47.2	29.7	68.5	49.3	32.1
36°	98.2	82.7	66.6	73.9	52.4	41.3	72.5	50.1	43.6
54°	96.4	76.9	61.1	75.0	46.9	38.8	75.5	51.5	39.3
72°	94.1	64.8	55.2	68.0	45.5	38.2	70.9	46.7	41.2
90°	91.5	63.1	54.6	68.2	43.9	38.5	69.4	43.0	39.6
108°	93.9	68.0	55.2	71.1	46.1	41.6	71.4	47.5	39.9
126°	97.5	76.9	59.1	72.9	48.1	44.9	73.8	50.9	43.7
144°	98.4	82.2	58.9	76.1	49.4	42.2	76.3	49.3	42.5
162°	95.8	75.4	48.8	70.4	43.6	33.4	68.3	45.4	35.7
180°	85.6	61.3	39.4	55.4	31.1	22.5	55.8	33.8	24.2
Mean	94.1	72.4	54.0	68.7	44.5	35.9	68.9	45.9	37.2
Total mean	73.5	49.7	50.7
Variances	14.9	61.2	70.8	46.0	34.7	51.1	44.7	29.8	40.9
Average variances	49.0	43.9	38.5