Fig. 1. PPF point pair features

Fig. 2. PRF point region features

Fig. 3. Features of different planes. (a) Point domain features on a plane; (b) intra-region normal vector on a plane; (c) point domain features on a surface; (d) intra-region normal vector on a surface

Fig. 4. PRF algorithm flowchart

Fig. 5. Establishing a local coordinate system

Fig. 6. Pose matching and voting process

Fig. 7. Test subject point cloud. (a) Wrench; (b) steel pipe; (c) T-shaped PVC pipe

Fig. 8. Influence of λ value on matching results. (a) Effect of λ on the number of clustering results of coarsely matched poses of different parts; (b) effect of λ on the RMSE of the estimation of coarse matching poses of different parts; (c) effect of λ on the MAE of the estimation of coarse matching poses of different parts

Fig. 9. Scene point cloud

Fig. 10. Rough matching algorithm accuracy. (a) RMSE; (b) MAE

Fig. 11. Mismatches problems in the test. (a) Reverse matching; (b) translational matching

Fig. 12. Accuracy of complete matching. (a) RMSE; (b) MAE

Fig. 13. Matching effect of PRF+AA-ICP

Table 1. Time and number of iterations used by different algorithms