Monocular VO Based on Deep Siamese Convolutional Neural Network

Table 2

Mean errors of each method on testing sequences.

Seq.	DSCNN-VO		CNN-VO		CNN-LSTM-VO		VISO2_M		VISO2_S
	trel	rrel	trel	rrel	trel	rrel	trel	rrel	trel	rrel
04	3.02	2.09	8.82	0.77	4.42	2.13	8.46	0.57	0.57	0.32
05	4.53	1.93	7.94	2.79	4.5	1.99	11.17	4.04	2.33	1.10
09	4.69	1.42	8.48	3.11	5.35	1.65	13.98	1.36	3.09	1.17
10	6.30	1.54	10.04	2.93	8.06	1.93	19.68	3.57	1.74	1.07
Mean	4.64	1.75	8.82	2.41	5.58	1.93	13.32	2.39	1.93	0.92

t_rel: average translational RMSE drift (%) at lengths of 100 m–800 m. r_rel: average rotational RMSE drift (o/100 m) on lengths of 100 m–800 m.