This repository is a modified implementation of Analyzing the Quality of Matched 3D Point Clouds of Objects by Igor Bogoslavskyi and Cyrill Stachniss. (link)
While the paper was developed for a dynamic real world set-up, this implementation is fine tuned for a lab environment and prioritizes accuracy over generalizability. The classification features mentioned in the paper have not been implemented. The idea here is to create a universal metric to compare quality of alignment in registered point clouds.
The implementation is located in pipeline.py, with support functions provided in quality_utils.py. This repository is still under development is prone to updates in the future.
The file pipeline.py takes in as input two unregistered point clouds and performs a two step alignment process. The initial global alignment makes use of RANSAC and FPFH features to coarsely align a down-sampled version of the point cloud, while fine alignment makes use of point to plane ICP algorithm to get a proper alignment.
This alignment is performed prior to calculating the quality metric. The probabilistic quality metric is a function of the several parameters used in developing the camera intrinsic and extrinsic parameters defined in the project_pcd_to_depth function in the utilities. Feel free to play around with them when tuning the metric with a ground truth. A value of around 0.8 to 0.9 is ideal. Anything less than 0.6 requires more tuning while anything above 1.0 indicates a need to relax the parameters.
The metric_development.ipynb notebook is a development playground and can be entirely ignored. Place all data files (point louds, meshes, etc.) in the data folder so that they can be references relatively without error.
The implementation has been developed and tested on a Windows system running Python 3.8. Open3d 0.18.0 is the desirable version required. Documentation for Open3d can be found here
- I. Bogoslavskyi and C. Stachniss, "Analyzing the quality of matched 3D point clouds of objects," 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Vancouver, BC, Canada, 2017, pp. 6685-6690, https://doi.org/10.1109/IROS.2017.8206584.
- Zhou, Q.-Y., Park, J., & Koltun, V. (2018). Open3D: A Modern Library for 3D Data Processing (Version 1). arXiv. https://doi.org/10.48550/ARXIV.1801.09847