crosstyan/zed-playground
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: implement 3D AABB overlap check for ICP registration

Browse Source
This commit is contained in:
crosstyan
2026-02-10 15:27:52 +00:00
parent 71b146bc72
commit adc38a441d
3 changed files with 49 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files
py_workspace/.sisyphus/notepads/full-icp-pipeline/learnings.md
+16 -19
View File
@@ -1,20 +1,17 @@
## 2026-02-10T09:45:00Z Session bootstrap
- Initial notepad created for full-icp-pipeline execution.
- Baseline code references verified in `aruco/icp_registration.py` and `refine_ground_plane.py`.
- Corrected success gate logic to `> 0`.
- Added INFO logging for all attempted ICP pairs.
- Ensured all pairs are stored in `metrics.per_pair_results`.
- Fixed overlap skip logging to use DEBUG level.
- Fixed syntax and indentation errors in `aruco/icp_registration.py` that were causing unreachable code and malformed control flow.
- Relaxed success gate to `metrics.num_cameras_optimized > 0`, allowing single-camera optimizations to be considered successful.
- Implemented comprehensive per-pair diagnostic logging: INFO for ICP results (fitness, RMSE, convergence) and DEBUG for overlap skips.
- Ensured all attempted ICP results are stored in `metrics.per_pair_results` for better downstream diagnostics.
- Updated `tests/test_icp_registration.py` to reflect the new success gate logic.
## Task 2: Point Extraction Functions
### Learnings
- Open3D's `remove_statistical_outlier` returns a tuple `(pcd, ind)`, where `ind` is the list of indices. We only need the point cloud.
- `estimate_normals` with `KDTreeSearchParamHybrid` is robust for mixed geometry (floor + walls).
- Hybrid extraction strategy:
1. Extract floor band (spatial filter).
2. Extract vertical points (normal-based filter: `abs(normal · floor_normal) < 0.3`).
3. Combine using boolean masks on the original point set to avoid duplicates.
- `extract_scene_points` provides a unified interface for different registration strategies (floor-only vs full-scene).
### Decisions
- Kept `extract_near_floor_band` as a standalone function for backward compatibility and as a helper for `extract_scene_points`.
- Used `mode='floor'` as default to match existing behavior.
- Implemented `preprocess_point_cloud` to encapsulate downsampling and SOR, making the pipeline cleaner.
- Added `region` field to `ICPConfig` to control the extraction mode in future tasks.
## Task 3: 3D AABB Overlap Check
- Implemented `compute_overlap_3d` in `aruco/icp_registration.py`.
- Added `overlap_mode` to `ICPConfig` (defaulting to "xz").
- Verified 3D overlap logic with new tests in `tests/test_icp_registration.py`.
- Confirmed that empty inputs return 0.0 volume.
- Confirmed that disjoint boxes return 0.0 volume.
- Confirmed that partial and full overlaps return correct hand-calculable volumes.