[3d] Port estimateTranslation2D into 5.x #28023
Merge with opencv_extra PR: opencv/opencv_extra#1290
This PR ports the newly added function `estimateTranslation2D` that was merged into 4.x under this PR: opencv/opencv#27950
The implementation is now adapted to the 5.x branch and integrates cleanly with the updated refinement API.
Specifically, this 5.x version makes use of the improved LevMarq interface to support optional non-linear refinement of the translation parameters.
Functionality, tests, and documentation follow the same design and behavior as the original 4.x PR.
Further details can be found in the linked PR above.
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [ ] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Feat #25150: Pose Graph MST initialization #27423
Implements an MST-based initialisation for pose graphs, as proposed in issue #25150. Both Prim’s and Kruskal’s algorithms were added to the 3D module. These receive a vector of node IDs and a vector of edges (each with source and target IDs and a weight), and return a vector with the resulting edges. These MST implementations treat edges as undirected internally, meaning users only need to provide one direction (A→B or B→A), and duplicates are handled automatically.
Additionally, a new pose graph initialisation method using MST (Prim) was implemented. It constructs the MST over the pose graph, then traverses it to reconstruct node poses. With this, users can call `poseGraph->initializePosesWithMST()` to create an initial solution for the pose graph problem.
A set of test cases validating the implementation was also included.
#### Notes
- The edge weight used in the MST for pose graphs is calculated as:
`weight = || translation || + λ * rotation_angle`,
where λ = 0.485 was determined empirically based on optimiser performance;
- Validated on [Sphere-a](https://lucacarlone.mit.edu/datasets/) pose graph, showing similar convergence behaviour with or without MST initialisation;
- Alternative weight formulas, such as the Mahalanobis distance formula, were also tested, but the current formula yielded better results.
#### Future Work
- Extend testing to more diverse pose graphs (currently limited to [Sphere-a](https://github.com/opencv/opencv_extra/blob/5.x/testdata/cv/rgbd/sphere_bignoise_vertex3.g2o) in opencv_extra)
- Explore adaptive tuning of the λ parameter for broader applicability.
Co-authored-by: @miguel1099
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [ ] There is a reference to the original bug report and related work
- [ ] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Move API focused C++ samples to snippets #25252
Clean Samples #25006
This PR removes 39 outdated C++ samples from the project, as part of an effort to keep the codebase clean and focused on current best practices.
OBJ and PLY loaders extention to support texture coordinates and difused colors #25221
### This PR changes
* Texture coordinates support added to `loadMesh()` and `saveMesh()`
* `loadMesh()` changes its behavior: all vertex attribute arrays (vertex coordinates, colors, normals, texture coordinates) now have the same size and same-index corresponce
- This makes sense for OBJ files where vertex attribute arrays are independent from each other and are randomly accessed when defining faces
- Looks like this behavior may also happen in some PLY files; however, it is not implemented until we encounter such files in a wild nature
- At the same time `loadPointCloud()` keeps its behavior and loads vertex attributes as they are given in the file
* PLY loader supports synonyms for the properties: `diffuse_red`, `diffuse_green` and `diffuse_blue` along with `red`, `green` and `blue`
* `std::vector<cv::Vec3i>` supported as an index array type
* Colors are loaded as [0, 1] floats instead of uchars
- Since colors are usually saved as floats, internal conversion to uchar at loading significantly drops accuracy
- Performing uchar conversion does not always makes sense and can be performed by a user if they needs it
* PLY loading fixed: wrong offset ruined x coordinate
* Python tests added for `loadPointCloud` and `loadMesh`
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
[GSoC] Update octree methods and create frames for PCC #23985
## PR for GSoC Point Cloud Compression
[Issue for GSoC 2023](https://github.com/opencv/opencv/issues/23624)
* We are **updating the Octree method create() by using OctreeKey**: Through voxelization, directly calculate the leaf nodes that the point cloud belongs to, and omit the judgment whether the point cloud is in the range when inserted. The index of the child node is calculated by bit operation.
* We are also **introducing a new header file pcc.h (Point Cloud Compression) with API framework**.
* We added tests for restoring point clouds from an octree.
* Currently, the features related to octree creation and point cloud compression are part of the internal API, which means they are not directly accessible to users. However, our plan for the future is to **include only the 'PointCloudCompression' class in the 'opencv2/3d.hpp' header file**. This will provide an interface for utilizing the point cloud compression functionality.
The previous PR of this was closed due to repo name conflicts, therefore we resubmitted in this PR.
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Triangle rasterization function #24459#24065 reopened since the previous one was automatically closed after rebase
Connected PR with ground truth data: [#1113@extra](https://github.com/opencv/opencv_extra/pull/1113)
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
PLY mesh support #24961
**Warning:** The PR changes exising API.
Fixes#24960
Connected PR: [#1145@extra](https://github.com/opencv/opencv_extra/pull/1145)
### Changes
* Adds faces loading from and saving to PLY files
* Fixes incorrect PLY loading (see issue)
* Adds per-vertex color loading / saving
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Merge usac to 5.x #24005
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Base branch is PR #23979. Merging PR #23078, 23900 and PR #23806 to 5.x
GSoC: Modified PLY reader to support color attribute read/write #23805
* Modified PLY reader to support color attribute read/write
* Fix bugs & Modified OBJ reader for color IO
* Replace with correct test file
* Fix I/O of property [w] in OBJ files
### Pull Request Readiness Checklist
**Merged with https://github.com/opencv/opencv_extra/pull/1075**
[Issue for GSoC 2023](https://github.com/opencv/opencv/issues/23624)
The ply loader in 3D module doesn't support color attribute reading. I modified that to support color attribute reading & writing for the color attribute compression as described in proposal.
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Fixes#22799
Replaces #21559 which was taken as a base
Connected PR in contrib: [#3388@contrib](https://github.com/opencv/opencv_contrib/pull/3388)
### Changes
OK, now this is more Odometry-related PR than Volume-related. Anyway,
* `Volume` class gets wrapped
* The same was done for helper classes like `VolumeSettings`, `OdometryFrame` and `OdometrySettings`
* `OdometryFrame` constructor signature changed to more convenient where depth goes on 1st place, RGB image on 2nd.
This works better for depth-only `Odometry` algorithms.
* `OdometryFrame` is checked for amount of pyramid layers inside `Odometry::compute()`
* `Odometry` was fully wrapped + more docs added
* Added Python tests for `Odometry`, `OdometryFrame` and `Volume`
* Added Python sample for `Volume`
* Minor fixes including better var names
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
### Changes
* Duplicated code removal in TSDF tests by implementing them with fixtures and GTest params
* e.g. separate OCL tests file removed
* as a result, more test cases are covered
* the same's done for perf tests
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Corresponding contrib PR: #3382@contrib
Changes
- Volume::raycast(): camera intrinsics can be explicitly passed to the function. If not, the ones from current volume settings are used
- getVolumeDimensions() renamed to getVolumeStrides() because they are strides actually
- TSDF tests: OpenCLStatusRevert and parametrized fixture
- ColorTSDF::integrate(): extra RGB projector is redundant, removed
- Minor changes
This PR contains:
- a new property enableGrowth which controls should the HashTSDF be extended during integration by adding new volume units or not
- a new method getBoundingBox which calculates the size of currently occupied data
- a set of tests to check that new functionality
- a fix for TSDF GPU reset (data is correctly zeroed now using floatToTsdf() function)
minor changes
There's a bug which appears when volume pose contains non-trivial rotation.
It results in wrong depth integration which can be observed during raycasting
or points/normals export.
- This PR fixes the bug for both CPU and OpenCL
- There is a reproducer for the bug
- The copy behavior of VolumeSettings fixed (now copy constructor creates a deep copy)
- Minor changes (e.g. unused vars removed)
Complement PR: #3366@contrib
Changes
OdometryFrame losts its getters: a user can provide data at construction stage only, pyramids and other generated data is read-only now
OdometryFrame is based on UMats: no TMat templates inside, CPU operations are done with UMat::getMat() method, chaining issues are solved ad-hoc
No more Odometry::createOdometryFrame() method, frames are compatible with all odometry algorithms
Normals computer is cached inside Odometry and exposed to API as well as its settings
Volume::raycast() won't return the result in OdometryFrame anymore
Added test for Odometry::prepareFrame*() & other test fixes
Minor code improvements
TODOs:
fix TODOs in code
lower acceptable accuracy errors
RgbdNormals: setMethod() removed as useless
RgbdNormals: tests + cross product, to be fixed
+ cross product
LINEMOD: diffThreshold param added + tests fixed
minor
diffThreshold fix
points3dToDepth16U fix
normals computer diffThreshold fix
random plane generation fixed + diffThreshold fix
Rendered normals test rewritten to GTest Params
random plane generation: scale
RGBD_Normals tests: thresholds tuned
Rendered normals tests: 64F support added
Random planes normal tests rewritten to GTest Params
LINEMOD and CrossProduct fix
SRI threshold raised
NormalsRandomPlanes: thresholds raised
assert on unknown alg; minor
fix
frame size reduced
TIFF replaced by YAML.GZ
depthTo3d test changed
cv::transform is used
fix warning
nanMask()
flipAxes()
absDotPixel() + forgotten code
helper functions removed
RGBDNormals: checkNormals() and compare LINEMOD's pts3d to depth input
Rendered: another criteria; thresholds; LINEMOD's pts3d to depth input comparison
thresholds raised a bit
SRI slightly optimized
assert change
normal tests refactored, parametrized, split
trailing namespace, thresholds raised
SRI caching optimized a lot
normal tests rewritten to fixture, no loop
minor
runCase() joined with testIt()
thresholds were put into GTest params
ternary operator
RgbdNormalsTest merged into NormalsRandomPlanes; RgbdPlanes moved closer to tests
normal test minor refactoring
plane finder tests refactored to GTest Params
skip tests
thresholds raised
plane test minor
plane tests: timers dropped, nPlanes put into GTest Params; refactoring
generated normals tests: minor refactoring
flipAxes() templated
rendered normals tests refactored: thresholds to GTest Params
CV_Error -> ASSERT_FALSE
Add normal estimation and region growing algorithm for point cloud
* Add normal estimation and region growing algorithm for point cloud
* 1.Modified documentation for normal estimation;2.Converted curvature in region growing to absolute values;3.Changed the data type of threshold from float to double;4.Fixed some bugs;
* Finished documentation
* Add tests for normal estimation. Test the normal and curvature of each point in the plane and sphere of the point cloud.
* Fix some warnings caused by to small numbers in test
* Change the test to calculate the average difference instead of comparing each normal and curvature
* Fixed the bugs found by testing
* Redesigned the interface and fixed problems:
1. Make the interface compatible with radius search.
2. Make region growing optionally sortable on results.
3. Modified the region growing interface.
4. Format reference.
5. Removed sphere test.
* Fix warnings
* Remove flann dependency
* Move the flann dependency to the corresponding test
GSoC module to save and load point cloud
* Add functionality to read point cloud data from files
* address issues found on review, add tests for mesh, refactor
* enable fail-safe execution and empty arrays as output
* Some improvements for point cloud io module
Co-authored-by: Julie Bareeva <julia.bareeva@xperience.ai>
Alexander Smorkalov
Ismail Abou Zeid
Alexander Smorkalov
Francisco Mónica
Souriya Trinh
WU Jia
Alexander Smorkalov
Gursimar Singh
Maksim Shabunin
Rostislav Vasilikhin
Zhangjie Chen
Maksym Ivashechkin
Alexander Alekhin
Yuhang Wang
Saratovtsev
Wanli Zhong
Klepikov DmitriiandJulie Bareeva
Ruan
Artem Saratovtsev