Visual Map Construction Using RGB-D Sensors for Image-Based Localization in Indoor Environments

Table 4

Localization errors of the query camera with different mapping methods.

Scenes	Methods of modeling	Average errors (meters)						Accuracy improvement (%)
		In direction		In direction		Of Euclidean distance
		Day	Night	Day	Night	Day	Night	Day	Night
Small room	Proposed method	0.4624	0.4152	0.3915	0.3689	0.6735	0.5940	—	—
	RGB-D ICP	0.5872	0.5490	0.4238	0.4065	0.7775	0.7208	13.38	17.59
	Classical ICP	0.6375	0.6304	0.4917	0.4375	0.8648	0.8182	22.12	27.40
Medium room	Proposed method	0.3956	0.4684	0.3956	0.3518	0.6885	0.6276	—	—
	RGB-D ICP	0.5974	0.5939	0.4376	0.4252	0.8019	0.7757	14.14	19.09
	Classical ICP	0.6951	0.6643	0.4636	0.4587	0.9042	0.8714	23.86	27.98
Large room	Proposed method	0.6656	0.6491	0.4802	0.4256	0.8727	0.8235	—	—
	RGB-D ICP	0.8070	0.7645	0.5468	0.5893	1.0429	1.0358	16.32	20.50
	Classical ICP	0.9372	0.9489	0.7466	0.6824	1.2848	1.2538	32.08	34.32
Corridor	Proposed method	0.5677	0.5890	0.4020	0.3974	0.7527	0.7575	—	—
	RGB-D ICP	0.6953	0.7041	0.4576	0.4539	0.8849	0.8888	14.94	14.77
	Classical ICP	0.7977	0.8084	0.6305	0.5918	1.0779	1.0708	30.17	29.26