Boundary Estimation from Point Clouds: Algorithms, Guarantees and Applications

Jeff Calder,Sangmin Park,Dejan Slepčev

DOI: https://doi.org/10.48550/arXiv.2111.03217

2022-05-12

Abstract:We investigate identifying the boundary of a domain from sample points in the domain. We introduce new estimators for the normal vector to the boundary, distance of a point to the boundary, and a test for whether a point lies within a boundary strip. The estimators can be efficiently computed and are more accurate than the ones present in the literature. We provide rigorous error estimates for the estimators. Furthermore we use the detected boundary points to solve boundary-value problems for PDE on point clouds. We prove error estimates for the Laplace and eikonal equations on point clouds. Finally we provide a range of numerical experiments illustrating the performance of our boundary estimators, applications to PDE on point clouds, and tests on image data sets.

Numerical Analysis,Statistics Theory,Machine Learning

What problem does this paper attempt to address?

### What problems does this paper attempt to solve?

This paper aims to address the challenge of identifying domain boundaries from point - cloud data. Specifically, the authors propose new algorithms and theoretical guarantees to estimate the following aspects:

1. **Boundary normal vectors**: Estimate the inner normal vector \(\nu(x)\) on the boundary.
2. **Distance to the boundary**: Estimate the distance \(d_\Omega(x)\) from an arbitrary point \(x\) to the boundary.
3. **Point detection within the boundary strip**: Provide a testing method to determine whether a point is located in the region \(\partial_\varepsilon\Omega\) within a distance \(\varepsilon\) from the boundary.

These problems are very important in many practical applications, such as:
- Solving partial differential equations (PDEs) on point clouds, especially when boundary conditions need to be imposed.
- Point - cloud - based applications such as anomaly detection and depth estimation.

### Research background and motivation

Traditional boundary detection methods have some shortcomings, such as high computational complexity, low accuracy, or lack of strict theoretical guarantees. Therefore, the authors propose a new method aiming at:
- **Efficiency**: The new algorithm can be calculated quickly.
- **Accuracy**: It is more accurate compared to the methods in the existing literature.
- **Theoretical guarantee**: Provide strict error estimates to ensure the reliability of the method.

### Main contributions

1. **Normal vector estimators**: Introduce first - order and second - order normal vector estimators \(\hat{\nu}_r\) and \(\hat{\nu}^2_r\).
2. **Distance estimators**: Define first - order and second - order distance estimators \(\hat{d}^1_r\) and \(\hat{d}^2_r\) respectively.
3. **Boundary strip test**: Design a statistical test \(\hat{T}_{\varepsilon,r}\) to identify points located within the boundary strip.
4. **Theoretical analysis**: Provide non - asymptotic error bounds and asymptotic error estimates to ensure the accuracy of the estimators.
5. **Application examples**: Demonstrate how to use these estimators to solve the Laplace equation and the Eikonal equation on point clouds and prove the convergence.

Through these methods, the authors not only improve the accuracy of boundary detection but also provide a reliable tool for solving PDEs on point clouds.