Home

DCReg Wiki

Language: English | 中文

DCReg is a Schur-complement-based framework for efficient degenerate LiDAR registration. It decouples rotation and translation observability, maps the raw eigenspace into physical motion axes, and stabilizes only the weak directions through a targeted preconditioned solve.

Start Here

• Algorithm Overview
• Installation and Quick Start
• Experiment Results
• Parking-Lot real-scene demo

Highlights

• Schur-complement-based spectral degeneracy detection removes misleading rotation-translation coupling before observability analysis.
• Physical-axis characterization maps weak modes to roll/pitch/yaw and x/y/z with aligned eigenvalues and contribution ratios.
• Targeted preconditioning stabilizes only the weak directions instead of damping the full coupled system.
• Lightweight runtime stack: Eigen + PCL, with optional TBB/OpenMP.

Repository Links

• Project Repository
• Paper on arXiv
• Video Demo

Notes

• Runtime code lives under DCReg/.
• The default sample input is DCReg/data/shifted_cylinder/measured_cloud_shifted_cylinder.pcd.
• The parking-lot source frame is bundled under DCReg/data/Parking-Lot-example/; download the large prior map from the README data link before running the real-scene demo.
• Next up: an open-source DCReg-based localization system that demonstrates integration into larger pipelines.