Fig.1

The ethnic building used for data collection"

Fig.2

Stereo imaging configuration"

Fig.3

Schematic diagram of symmetrical stereo configurations for data collection"

Fig.4

Implementation of data collection"

Fig.5

Framework of the skeletal camera network generation with a data driven approach in analyzing stereo configuration for 3D geometry reconstruction of terrestrial images"

Fig.6

The TCN of an example with 10 images"

Fig.7

Schematic diagram of control points distribution. (L1—L3 show three diagrams of centralized distribution; L4 and L5 are two diagrams of edge distribution; L6 and L7 are two diagrams of even distribution)"

Fig.8

Positioning errors of stereo matching with different number of control points"

Fig.9

Positioning errors of X, Y, Z coordinates on the achieved geometry from stereo pairs, where X coordinate is parallel to the image shooting direction"

Fig.10

Positioning error of stereo pairs with different base-height ratio values"

Fig.11

Positioning error of reconstructed 3D geometry from stereo pairs with different intersection angle values"

Fig.12

Configuration of terrestrial images (where the used images are marked in yellow)"

Fig.13

TCNs of the three datasets"

Fig.14

Weighted TCNs"

Fig.15

SCNs of the three datasets"

Fig.16

Comparison of the number of edges in the networks of DN and SCN used for the typical SfM and the SCN-SfM, respectively, for the three datasets with 10, 96, and 650 images"

Fig.17

Comparison of the time cost in the triangulation stage with the typical SfM and the SCN-SfM methods for the three datasets with 10, 96, and 650 images"

Fig.18

Sparse point clouds of the reconstructed geometry for the three datasets"

Fig.19

Dense point clouds of the reconstructed geometry for the three datasets"

Fig.20

Geometric accuracy of the achieved point clouds on the three datasets by the typical SfM and the proposed SCN-SfM methods"