Multi‑Instance Point Cloud Pose Estimation Method Based on Gaussian‐Weighted Voting Strategy

Table 1. Accuracy comparison results of the best pose estimation of workpiece
View table
Table 1. Accuracy comparison results of the best pose estimation of workpiece
Workpiece RE /（°） TE /mm
PPF Proposed algorithm PPF Proposed algorithm
Brick 2.70 1.65 2.97 1.81
Gear 1.27 0.01 0.87 0.23
Linkage 5.98 3.57 2.55 2.36

Table 2. Pose estimation evaluation results of Romain dataset
View table
Table 2. Pose estimation evaluation results of Romain dataset
Algorithm MR MP MF
PPF 60.40 60.40 60.40
3D Hough 6.18 7.21 6.15
Proposed algorithm 72.94 83.22 76.96

Table 3. Pose estimation evaluation results of ROBI dataset
View table
Table 3. Pose estimation evaluation results of ROBI dataset
Algorithm MR MP MF
Line 2D 30.96 3.22 4.90
AAE 26.15 5.11 8.37
PPF 6.41 4.58 4.26
3D Hough — — —
Proposed algorithm 20.04 34.25 23.76

Table 4. Evaluation results of pose estimation of proposed algorithm under real point cloud dataset
View table
Table 4. Evaluation results of pose estimation of proposed algorithm under real point cloud dataset
Sequence RMSE /mm MR /% MP /% MF /%
1 2.84 37.50 37.50 37.50
2 1.13 41.18 53.85 46.67
3 2.16 66.67 85.71 75.00
4 2.58 70.00 77.78 73.69
5 1.80 90.00 90.00 90.00

Table 5. Success rate of robotic arm grasping in real scenes
View table
Table 5. Success rate of robotic arm grasping in real scenes
Algorithm
Experiment
No.
Grasping
success rate
Average grasping
success rate
PPF 1 80 60
2 70
3 45
4 65
5 40
Proposed
algorithm
1 100 93
2 85
3 85
4 100
5 95

Tools

Get Citation

Copy Citation Text

Ying Zhang, Hongzhi Du, Yunbo Hu, Yanbiao Sun, Jigui Zhu. Multi‑Instance Point Cloud Pose Estimation Method Based on Gaussian‐Weighted Voting Strategy[J]. Acta Optica Sinica, 2025, 45(5): 0515001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites