AWCD: An Efficient Point Cloud Processing Approach via Wasserstein Curvature

Yihao Luo,Ailing Yang,Fupeng Sun,Huafei Sun

DOI: https://doi.org/10.48550/arXiv.2105.04402

2021-05-11

Abstract:In this paper, we introduce the adaptive Wasserstein curvature denoising (AWCD), an original processing approach for point cloud data. By collecting curvatures information from Wasserstein distance, AWCD consider more precise structures of data and preserves stability and effectiveness even for data with noise in high density. This paper contains some theoretical analysis about the Wasserstein curvature and the complete algorithm of AWCD. In addition, we design digital experiments to show the denoising effect of AWCD. According to comparison results, we present the advantages of AWCD against traditional algorithms.

Machine Learning,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

This paper attempts to solve the denoising problem of point cloud data under high - density noise pollution. Traditional point cloud denoising methods (such as Radius - based Outlier Removal (ROR) and Statistical Outlier Removal (SOR)) are not effective in dealing with high - density noise and cannot effectively remove noise and retain real information.

To solve this problem, the paper proposes a new point cloud processing method - Adaptive Wasserstein Curvature Denoising (AWCD). This method can more accurately describe the data structure by introducing Wasserstein distance and curvature information, and can maintain stability and effectiveness in the case of high - density noise.

Specifically, the main contributions of the AWCD method include:

1. **Introducing Wasserstein Curvature**: Use the Wasserstein distance to calculate the curvature information of point cloud data. These curvatures can reflect the relationship between points and their neighborhoods, so as to distinguish real information from noise.
2. **Adaptability**: By constructing a curvature histogram, AWCD can automatically determine a "marked curvature" to distinguish real information from noise without requiring too many manual parameters.
3. **High Efficiency**: The experimental results show that AWCD performs well on data sets of different scales and noise densities. Especially for data sets with high - density noise, its denoising effect is significantly better than that of traditional methods.

In summary, this paper aims to improve the denoising method of point cloud data by introducing Wasserstein curvature, make it more adaptable to high - density noise situations, and improve the denoising effect.