Fig. 1. Proposed data alignment strategy for multi-view structured light measurement

Fig. 2. Pose estimation based on YOLO network

Fig. 3. Experimental results. (a) Setup; (b) training loss and testing accuracy; (c) translation and rotation error; (d) true pose determination; (e) pose estimation visualization

Fig. 4. More cases presentation of object pose estimation. (a) Sphere; (b) pyramid; (c) pillars; (d) elbow

Fig. 5. Single-view structured light reconstruction based on the proposed system. (a) Projection image; (b) wrapping phase; (c) absolute phase; (d) 3D point cloud

Fig. 6. Point cloud splicing using estimated pose. (a)(d) Projection images in two views; (b)(e) pose estimation results; (c)(f) point clouds in two views; (g) data splicing result; (h) zoom-in view of box in Fig. (g)

Fig. 7. Data registration with estimated pose. (a)(d)Two-view registration of pillars and recess, with deep learning-based pose estimation; (b)(e) global refinement using ICP algorithm based on rough rigid transformation; (c)(f) error distributions of the final registration of fused data in Figs. (b), (e) and CAD model, where the error is determined by the point-to-model distance

Table 1. Error computation of pose estimation

Table 2. Error comparison of data fusion using markers and deep learning