Head Pose Estimation with Improved Random Regression Forests

Table 1

The accuracy and MAE for different methods on the Pointing’04 database.

Method	MAE		Accuracy
	Yaw	Pitch	Yaw	Pitch
WRF [7]	7.5°	7.8°	80.15%	78.23%
VRF + LDA [17]	11.05°	N/A	66.95%	N/A
SRF [18]	9.6°	13.9°	N/A	N/A
LPLS [29]	11.29°	10.52°	45.57%	58.70%
KPLS [29]	6.56°	6.61°	67.36%	80.36%
MGD [31]	6.9°	8.0°	64.51%	62.72%
MLD [9]	4.24°	2.69°	N/A	N/A
WSRF (Gabor)	6.32°	5.92°	84.14%	85.41%
WSRF (LBP)	6.23°	5.65°	83.01%	85.78%
WSRF (HOG)	5.21°	4.06°	87.32%	89.41%
WSRF (intensity)	6.54°	6.41°	83.68%	84.86%

N/A denotes that the measure is not available in the paper where the method was proposed (same for the other tables in this paper). For the method SRF, 14 subjects are used to train the random regression forests while the rest subjects are applied for test.