Vision-Based Satellite Recognition and Pose Estimation Using Gaussian Process Regression

Table 5

Pose estimation results on the 1D subset of BUAA-SID 1.5. Results are in bold if the proposed method performs the best and in italic if the second best.

Representation	Method	MAE (°)	(%)	(%)	(%)
BI	MCGPR	4.83	53.44	70.11	88.06
	MCGPR-ALL	6.13	33.67	53.56	78.72
	GPR	11.4	22.00	38.33	61.00
	GPR-ALL	14.46	10.28	21.33	45.61
	Kernel method [19]	4.84	53.33	70.17	88.22
DT	MCGPR	6.78	48.94	63.22	78.72
	MCGPR-ALL	8.82	20.67	39.22	67.50
	GPR	14.8	18.39	32.50	52.33
	GPR-ALL	20.60	5.83	11.39	28.06
	Kernel method [19]	6.07	53.67	68.33	82.56
HU	MCGPR	57.3	4.28	8.17	17.50
	MCGPR-ALL	75.21	1.33	2.33	6.56
	GPR	69.5	1.06	2.39	5.17
	GPR-ALL	83.94	0.67	1.33	2.89
	kernel method [19]	80.37	1.22	2.61	6.44
FD	MCGPR	23.2	8.44	16.94	35.78
	MCGPR-ALL	34.81	3.50	7.11	16.89
	GPR	38.9	3.17	6.78	14.94
	GPR-ALL	56.54	0.94	2.39	6.22
	Kernel method [19]	27.09	8.50	16.28	32.83
GI	MCGPR	0.29	93.06	98.00	99.72
	MCGPR-ALL	0.43	90.44	96.56	99.50
	GPR	3.02	52.33	72.44	90.67
	GPR-ALL	3.84	38.89	62.00	86.72
	Kernel method [19]	0.29	93.06	97.94	99.72
	ResNet [46]	4.5206	17.22	31.67	74.44
	DenseNet [47]	22.41	2.22	6.11	15.56
HOG	MCGPR	5.32	43.61	64.72	82.83
	MCGPR-ALL	6.04	27.44	48.61	77
	GPR	15.5	9.67	18.89	42.06
	GPR-ALL	16.51	6.61	13.67	35.11
	Kernel method [19]	5.36	43.50	64.72	82.94