Infrared-based facial points tracking and action units detection in context of car driving simulator

EURASIP Journal on Image and Video Processing

Table 3 Precision and tracking fragmentation of the different facial points tracking methods for the SPEED-Q dataset

Rad	MC	TM	C ₁	C ₂	C ₃	C ₄	C ₅	C ₆	C ₇	C ₈	C ₉	C ₁₀	C ₁₁	C ₁₂	C ₁₃	C ₁₄	C ₁₅	C ₁₆	C ₁₇	C ₁₈	$\overset{Â¯}{MC}$
10%	P%	PF	49	51	71	70	72	48	52	72	70	69	62	65	63	67	69	63	66	67	63
		AAM	53	52	74	72	70	55	54	67	69	71	65	66	63	71	68	64	67	70	65
		HCPF-AAM	75	76	87	88	84	74	72	89	86	83	82	84	83	93	95	92	96	89	84
20%	P%	PF	42	41	65	66	63	39	41	62	64	61	53	56	52	58	53	59	57	64	55
		AAM	45	46	68	69	67	43	44	66	68	65	57	53	52	64	66	62	65	65	59
		HCPF-AAM	72	70	86	84	84	71	70	85	82	80	81	82	80	91	92	92	93	87	82
50%	P%	PF	36	37	60	62	59	35	36	61	59	60	48	50	46	52	50	49	51	57	50
		AAM	40	39	62	66	62	37	38	61	58	55	52	51	50	59	58	52	55	60	53
		HCPF-AAM	67	69	85	84	77	66	64	84	80	78	78	79	77	87	88	86	89	85	79
–	TF%	PF	27	26	22	24	23	28	27	23	25	22	25	24	24	22	23	21	20	21	24
		AAM	25	27	21	20	19	24	22	19	20	23	22	23	21	20	22	17	19	20	21
		HCPF-AAM	14	12	8	9	9	13	11	8	7	9	10	12	11	8	6	7	6	6	9

Rad, radius circle size; TM, tracking method; MC, metric; $\overset{Â¯}{MC}$ , average of metric values for all facial points; C₁ to C₁₈, facial points shown in Figure 2.