3D Pose Matching Estimation Configuration Guide
About 1082 wordsAbout 4 min
3D pose matching estimation runs after 3D preprocessing and before Pick Point processing. It calculates instance poses based on the target type, Point Cloud features, and Task type. This page helps you choose a pose estimation method and find the configuration guide for each Task.
Configuration principles
Choose a pose estimation method based on the target geometry and Task type. Then check that the template, instance Point Cloud, and coordinate frame meet the method’s requirements. When tuning parameters, consider matching accuracy, success rate, and Cycle time together.
Choose a Pose Estimation Method
| Task type or target feature | Configuration guide |
|---|---|
| Bag or carton depalletizing | Bag Pose Generation and Carton Pose Estimation Configuration Guide The same set of parameters is called “Bag Pose Generation” for bag Tasks and “Carton Pose Estimation” for carton Tasks. It generates depalletizing poses using rectangle fitting, coordinate frames, and Pick Point neighborhood parameters. |
| Loading and unloading of neatly arranged general targets, picking randomly arranged general targets, or positioning and assembly of general targets | General Pose Estimation Configuration Guide Estimates the poses of general targets through Coarse Matching of keypoints and Fine matching of Point Clouds. |
| Loading and unloading of neatly arranged planar targets, picking randomly arranged planar targets, or positioning and assembly of planar targets | Point Cloud Matching Pose Estimation Configuration Guide Performs Coarse Matching and Fine matching between the template Point Cloud and instance Point Cloud. Suitable for planar targets whose identification depends on Point Cloud geometry. |
| Loading and unloading of neatly arranged planar targets that requires parallel matching | Point Cloud Matching Pose Estimation (Parallel) Configuration Guide Uses image or depth features to assist Point Cloud matching. Suitable for Tasks that must balance processing multiple targets with Cycle time. |
| Image matching Tasks for planar targets | Image Matching Pose Estimation Configuration Guide Uses image features for Coarse Matching, then refines the pose through Point Cloud Fine matching. |
| Targets with circular surfaces | Basic Shape Pose Estimation Configuration Guide Fits a circular surface and refines the pose using the template Point Cloud. |
| Cylindrical targets | Cylinder Pose Estimation Configuration Guide Fits a cylinder and standardizes its axis and pose orientation. |
| Quadrilateral targets | Quadrilateral Pose Estimation Configuration Guide Fits a quadrilateral plane and constrains the pose using dimensions and fit quality. |
General Tuning and Validation Principles
- Check that the instance Point Cloud is complete and that the template Point Cloud matches the current target before adjusting matching parameters.
- Validate the Coarse Matching result before adjusting Fine matching. If the pose from Coarse Matching is incorrect, Fine matching usually cannot produce a stable result.
- Adjust only one parameter at a time, and record the matching success rate, pose error, and computation time.
- Lowering a Threshold or narrowing the search range may improve accuracy but may also reduce the success rate. Expanding the search range or increasing sampling usually increases Cycle time.
- Repeat validation with field data from different positions, orientations, and batches so the parameters do not work for only a single frame.
Related Documents
Clean the Point Cloud and prepare geometric features before pose estimation
Adjust the position and orientation of Pick Points based on pose estimation results