EURASIP Journal on Image and Video Processing

Table 2 Evaluation of separate descriptors and all possible combinations of them on the test set

From: Driver head pose estimation using efficient descriptor fusion

	3 classes for pitch-SVM			3 classes for yaw-SVM
Descriptor	Acc.	κ	Time	Acc.	κ	Time
SF	87.2	0.80	0.05	94.3	0.91	0.03
HOG	85.6	0.77	0.01	94	0.90	0.01
SURF	83.8	0.75	0.04	93.7	0.90	0.04
Haar	85.9	0.78	0.2	93	0.89	0.2
(SF, HOG)	89.3	0.83	0.07	96.3	0.94	0.06
(SF, SURF)	89.0	0.83	0.13	95.3	0.92	0.12
(SF, Haar)	86.7	0.79	0.5	94.7	0.91	0.47
(HOG, Haar)	88.8	0.82	0.24	94.5	0.91	0.21
(HOG, SURF)	87.2	0.80	0.06	94.9	0.92	0.06
(SURF, Haar)	87.3	0.80	0.35	94.6	0.91	0.32
(SF, HOG, SURF)	89.1	0.83	0.15	95.4	0.93	0.11
(SF, HOG, Haar)	85.6	0.77	0.28	95.1	0.92	0.29
(SF, SURF, Haar)	77.9	0.64	0.53	92.3	0.88	0.48
(HOG, SURF, Haar)	87.8	0.81	0.19	94.9	0.92	0.17
(SF, HOG, SURF, Haar)	87.5	0.80	0.53	94.9	0.91	0.52

Italic values in Table 2: Best results obtained by all possible combinations of one, two, three and four descriptors

Back to article page