Deep Algebraic Fitting for Multiple Circle Primitives Extraction from Raw Point Clouds
Organizations: Nanjing University of Aeronautics and Astronautics · University of Oxford
Abstract
The shape of circle is one of fundamental geometric primitives of man-made engineering objects. Thus, extraction of circles from scanned point clouds is a quite important task in 3D geometry data processing. However, existing circle extraction methods either are sensitive to the quality of raw point clouds when classifying circle-boundary points, or require well-designed fitting functions when regressing circle parameters. To relieve the challenges, we propose an end-to-end Point Cloud Circle Algebraic Fitting Network (Circle-Net) based on a synergy of deep circle-boundary point feature learning and weighted algebraic fitting. First, we design a circle-boundary learning module, which considers local and global neighboring contexts of each point, to detect all potential circle-boundary points. Second, we develop a deep feature based circle parameter learning module for weighted algebraic fitting, without designing any weight metric, to avoid the influence of outliers during fitting. Unlike most of the cutting-edge circle extraction wisdoms, the proposed classification-and-fitting modules are originally co-trained with a comprehensive loss to enhance the quality of extracted circles.Comparisons on the established dataset and real-scanned point clouds exhibit clear improvements of Circle-Net over SOTAs in terms of both noise-robustness and extraction accuracy. We will release our code, model, and data for both training and evaluation on GitHub upon publication.
Figures & tables
| Description | Operation |
| Shape | Thirty multi-circle CAD models, including both curved and flat thin-walled planes |
| Radius | The largest radius is about five times the smallest |
| Depth | Between one-tenth and four times the radius |
| Scanning direction | Between fifteen and ninety degrees |
| Noise | 0.1%, 0.5%, and 1.0% Gaussian noise |
| Density | Three different sampling resolutions |
| Method | Normal-based | Clustering-based | Geometry-based | EC-NET | PIE-NET | Ours |
| 39.04 | 40.38 | 33.93 | 32.60 | 77.95 | 86.06 | |
| 70.10 | 22.15 | 69.81 | 26.82 | 69.79 | 77.62 | |
| 49.21 | 28.61 | 45.66 | 29.43 | 73.64 | 81.58 |
| Noise level | 0.1% noise | 0.5% noise | 1% noise | ||||||
| Method | |||||||||
| Hyper | 0.465328 | 0.296654 | 0.089543 | 0.38116 | 0.31696 | 0.103115 | 0.667738 | 0.271244 | 0.076074 |
| LLS | 0.466897 | 0.296752 | 0.089596 | 0.385131 | 0.321495 | 0.105989 | 0.675546 | 0.279059 | 0.08018 |
| RANSAC | 0.438734 | 0.309176 | 0.103949 | 0.454232 | 0.377379 | 0.172154 | 0.809487 | 0.281098 | 0.108321 |
| RLTS | 0.469162 | 0.278478 | 0.080783 | 0.50883 | 0.239431 | 0.092588 | 0.707097 | 0.252474 | 0.075499 |
| PIE-NET | 0.649781 | 0.312905 | 0.121466 | 0.67379 | 0.312294 | 0.155506 | 0.783924 | 0.268966 | 0.207283 |
| Method | |||
| 80.47 | 71.08 | 75.49 | |
| 84.45 | 76.91 | 80.51 | |
| 78.75 | 72.55 | 75.52 | |
| 79.40 | 74.37 | 76.80 | |
| 83.18 | 75.95 | 79.40 | |
| 86.03 | 77.62 | 81.58 |
| Noise level | 0.1% noise | 0.5% noise | 1% noise | ||||||
| Method | |||||||||
| 0.47372 | 0.291626 | 0.086550 | 0.378543 | 0.284565 | 0.083755 | 0.683557 | 0.203256 | 0.043517 | |
| 0.441876 | 0.313686 | 0.100596 | 0.361986 | 0.330956 | 0.111946 | 0.668966 | 0.277501 | 0.079145 | |
| 0.502384 | 0.283151 | 0.082876 | 0.484761 | 0.219477 | 0.069472 | 0.910986 | 0.022152 | 0.103095 | |
| 0.448136 | 0.266749 | 0.083464 | 0.364695 | 0.235270 | 0.073410 | 0.691933 | 0.086769 | 0.033559 | |
| 0.432098 | 0.127128 | 0.018671 | 0.42052 | 0.093475 | 0.01345 | 0.680641 | 0.018848 | 0.007759 | |
| Method | Hyper | LLS | RANSAC | RLTS | PIE-NET | PolyWorks | Ours |
| 0.027201 | 0.028264 | 0.052397 | 0.068493 | 0.060182 | 0.0267 | 0.016884 | |
| 0.001113 | 0.001166 | 0.005844 | 0.009348 | 0.006742 | 0.001045 | 0.000530 |
Parameter Estimation
Parameter estimation studies how to infer unknown quantities in statistical, physical, and machine-learning models from observed data.
See all 156 papers