Fig.1

Comparison of different geo-registration effects of outdoor ARGIS"

Fig.2

Effects of vision-aided pose optimization and geo-registration accuracy improvement"

Fig.3

The Vision-Aided Localization and Geo-Registration (VALGR) framework"

Fig.4

Image semantic segmentation and building area extraction"

Fig.5

Line segments extraction from building facades and vanishing points estimation"

Fig.6

Estimation of the boundary of vertical edges corresponding to the left and right corner of the building"

Fig.7

Recognition and extraction of the left, the middle and right corner vertical edge of the building"

Fig.8

Horizontal distribution of the three vertical corner edges and their corresponding corners on the 2D map"

Fig.9

Initial 2D camera pose and the 2D pose hypotheses"

Fig.10

Screening to get the optimal 2D pose solution"

Fig.11

The test scenarios and results of our experiments: for each triplet image in the vertical direction. (The top row demonstrates the three corner edges automatically extracted from the original images (after the rolling angle correction); The middle row demonstrates the semantic segmentation results of original images; And the third row demonstrates the sensor’s initial 2D pose and the optimized 2D pose.)"

Tab.1

The 3DOF translation accuracy of our method compared with the initial position acquired by sensorsm"

Tab.2

The 3DOF attitude accuracy of our method compared with the initial attitude acquired by sensors(°)"

Fig.12

Improvement of horizontal translation accuracy of each scene"

Fig.13

Improvement of yaw angle accuracy for each set of scenes"

Fig.14

Schematic diagram for quantitative registration accuracy evaluation"

Tab.3

The geo-registration accuracy of our method compared with the initial result acquired by sensors"

Fig.15

The geo-registration accuracy before and after optimization for each scene"

Fig.16

Geo-registration results in scene 2 before and after optimization"

Fig.17

Geo-registration results in scene 4 before and after optimization"