Overview

Robust pose estimation of objects with feature ambiguity based on point cloud.

Youtube video, If you cannot link to youtube, try bilibili.

Getting Started

Prerequisites

• Eigen3

• Sophus

• PCL

• OpenCV: modify the OpenCV_DIR to your path

• OpenMP

• Boost

• TdLib: modify the TdLib_DIR to your path

• nlohmann_json

Usage

• add model and generate the offline data, skip to add_model/README.md to know how to.
• capture the object point cloud in the scene add save it in the data in the pcd format and in meters unit.
• modify the config.ini. The main item you need to modify is model_name,scene_name, cad_models_path,model_data_path,
• in the root directory

  mkdir build
  cd build
  cmake ..
  make -j4

• ./globalPipeline

Citation

@ARTICLE{10261416,
  author={Li, Hai and Zeng, Qingfu and Zhuang, Tingda and Huang, Yanjiang and Zhang, Xianmin},
  journal={IEEE Sensors Journal}, 
  title={Accurate Pose Estimation of the Texture-Less Objects With Known CAD Models via Point Cloud Matching}, 
  year={2023},
  volume={23},
  number={21},
  pages={26259-26268},
  keywords={Point cloud compression;Pose estimation;Solid modeling;Feature extraction;Three-dimensional displays;Genetic algorithms;Data models;Feature ambiguity;iterative closest point (ICP);point cloud matching;pose estimation;texture-less objects},
  doi={10.1109/JSEN.2023.3316457}}

Acknowledgment

add_model part is modified from the object_identification_localization project.