3D map (color point clouds) that is globally consistent with buildings displayed on a publicly available map (gray point clouds).

Contexts in source publication

Context 1

... for alignment and has a user interface (UI) to confirm or modify the 3D alignment results. The user can check and fix the anchor pose from the anchor pose candidates and perform pose-graph optimization by checking the alignment results on the user interface. The optimization eliminates cumulative errors and generates a globally consistent 3D map. Fig. 1 shows a globally consistent 3D map generated by the proposed framework. The 3D map (color point clouds) coincides with the publicly available maps (gray point clouds). ...

Context 2

... process of alignment and optimization is shown in Algorithm 1. end while 20: end for 21: return X * When the anchor pose was located at a long distance from the building, as shown in Fig. 6 (a), the user pushes the ''Pose correction button.'' This process can correct the current anchor pose manually in the UI (Algorithm 1 line7-9). The user can freely specify the 6DoF pose of the anchor. ...

Context 3

... of them. The pose correction was necessary when the SLAM accumulation error was large and the ICP alignment did not converge. The user corrected the anchor pose by performing the ICP again with a manually input initial position aligned with the wall, such as in Fig. 6(b). The rejections occurred when the detected plane was not registered in OSM (Fig. 10(a)) or when the detected plane was not a plane of a building (Fig. 10(b)). These pose candidates were eliminated manually. The average number of operations by the user is approximately five per kilometer: two pose corrections and three rejections. These results show that a globally consistent 3D map can be generated with a small number of ...

Context 4

... error was large and the ICP alignment did not converge. The user corrected the anchor pose by performing the ICP again with a manually input initial position aligned with the wall, such as in Fig. 6(b). The rejections occurred when the detected plane was not registered in OSM (Fig. 10(a)) or when the detected plane was not a plane of a building (Fig. 10(b)). These pose candidates were eliminated manually. The average number of operations by the user is approximately five per kilometer: two pose corrections and three rejections. These results show that a globally consistent 3D map can be generated with a small number of user ...

Context 5

... estimation result is incorrect. The proposed method achieves the highest results. In particular, because the proposed method considers whether the scan information includes the wall surface, the anchor pose can be estimated with higher results than in the beside-buildings method, which simply uses the pose graph at the side of the wall surface. Fig. 11 shows the trajectory before and after optimization. The light blue arrow represents the selected anchor poses, and the yellow line represents the building wall near the reference trajectory to be detected. The results show that the proposed method can select an effective anchor pose near the wall surface. indicates that our framework ...

Context 6

... represents the selected anchor poses, and the yellow line represents the building wall near the reference trajectory to be detected. The results show that the proposed method can select an effective anchor pose near the wall surface. indicates that our framework is sufficiently accurate and can use publicly available maps for the navigation task. Fig. 13 shows the pose graphs for route 5; Fig. 13 (a) is a top view of the pose graphs and Fig. 13 (b) is a longitudinal profile. The other SLAMs show several deviations in their final positions because of a few degrees of deviation in the curve position. The NDT SLAM seems to overlap the trajectory of the reference in the top view, but the ...

Context 7

... yellow line represents the building wall near the reference trajectory to be detected. The results show that the proposed method can select an effective anchor pose near the wall surface. indicates that our framework is sufficiently accurate and can use publicly available maps for the navigation task. Fig. 13 shows the pose graphs for route 5; Fig. 13 (a) is a top view of the pose graphs and Fig. 13 (b) is a longitudinal profile. The other SLAMs show several deviations in their final positions because of a few degrees of deviation in the curve position. The NDT SLAM seems to overlap the trajectory of the reference in the top view, but the longitudinal profile shows that it is greatly ...

Context 8

... the reference trajectory to be detected. The results show that the proposed method can select an effective anchor pose near the wall surface. indicates that our framework is sufficiently accurate and can use publicly available maps for the navigation task. Fig. 13 shows the pose graphs for route 5; Fig. 13 (a) is a top view of the pose graphs and Fig. 13 (b) is a longitudinal profile. The other SLAMs show several deviations in their final positions because of a few degrees of deviation in the curve position. The NDT SLAM seems to overlap the trajectory of the reference in the top view, but the longitudinal profile shows that it is greatly deviated in the vertical direction. When the ...

Context 9

... results of optimized pose graph are also included in the released data. Fig.14(a) also shows the manual alignment of the LeGO-LOAM trajectory. ...

Context 10

... is an example where the loop closure is successful but the result does not correspond to the published map. Fig.14(b) shows the trajectory optimized by the proposed method. ...