Model-Matching

This software tool could be used to obtain robust 6d poses of objects with 3d point cloud models in the presence of noisy segmentation data

Robust 6D Object Pose Estimation with Stochastic Congruent Sets (pdf)(website)

By Chaitanya Mitash, Abdeslam Boularias, Kostas Bekris (Rutgers University).

In Proceedings of British Machine Vision Conference (BMVC), Newcastle, England, UK, 2018

Citing

To cite the work:

@article{mitash2018robust,
  title={Robust 6D object pose estimation with stochastic congruent sets},
  author={Mitash, Chaitanya and Boularias, Abdeslam and Bekris, Kostas},
  journal={arXiv preprint arXiv:1805.06324},
  year={2018}
}

Dependency

1. OpenCV
2. PCL

Installation

1. Download the repository.
2. mkdir build
3. cd build
4. cmake ../
5. make

Inputs

1. RGB and depth images
2. Per-pixel object class probability (scaled to range 0-10000 and stored as uint16). Can be set as a constant mask if probability is not available.

Outputs

1. best_pose_candidate_{object_name} 6D pose of the object (3 rows of the transformation matrix) stored in row-major order.
2. best_pose.ply and scene.ply visualization of the transformed object model and the scene.

Running the first example

1. Set the repo_path in files model_preprocess.cpp and stocs_match_one_object.cpp
2. Preprocess the 3d model ./build/model_preprocess "024_bowl"
3. Run pose estimation ./build/stocs_single "{path_to_repo}/examples/ycb/" "024_bowl"

Running on Packed-dataset

1. Change the following parameters in the file model_preprocess.cpp float voxel_size = 0.005;

2. Change the following parameters in the file stocs_match_one_object.cpp

std::vector<float> cam_intrinsics = {615.957763671875, 308.1098937988281, 615.9578247070312, 246.33352661132812};
float depth_scale = 1/8000.0f;

Running on Linemod

1. Change the following parameters in the file model_preprocess.cpp

float voxel_size = 10;
float normal_radius = 5;
float model_scale = 1.0f/1000;

2. Change the following parameters in the file stocs_match_one_object.cpp

std::vector<float> cam_intrinsics = {572.4114, 325.2611, 573.57043, 242.04899};
float depth_scale = 1/1000.0f;