Weighted Multi-Projection: 3d Point Cloud Denoising With Tangent Planes

Abstract

As a collection of 3D points sampled from surfaces of objects, a 3D point cloud is widely used in robotics, autonomous driving and augmented reality. Due to the physical limitations of 3D sensing devices, 3D point clouds are usually noisy, which influences subsequent computations, such as surface reconstruction, recognition and many others. To denoise a 3D point cloud, we present a novel algorithm, called weighted multi-projection. Compared to many previous works on denoising, instead of directly smoothing the coordinates of 3D points, we use a two-fold smoothing: We first estimate a local tangent plane at each 3D point and then reconstruct each 3D point by weighted averaging of its projections on multiple tangent planes. We also provide the theoretical analysis for the surface normal estimation and achieve a tighter bound than in a previous work. We validate the empirical performance on the dataset of ShapeNetCore and show that weighted multi-projection outperforms its competitors in all nine classes.

Feature Graph Learning for 3D Point Cloud Denoising

“…2) Objective Comparison: We measure the quality of denoised results for Benchmark models by the Mean Squared Error (MSE) and Signal-to-Noise Ratio (SNR) between each denoised point cloud and the ground truth as in [62]. Numerical results are listed in Table I and Table II, respectively.…”

“…We note that this is one possible graph connectivity among many. For example, one can in addition enable intra-patch filtering by drawing connections among points in the same patch [61], [62], resulting in a denser graph. For simplicity, we employ the most basic graph connectivity given inter-patch similarities.…”

“…Specifically, we consider two features: Cartesian coordinates and normals. As done in [61], [62], surface normals are adopted to promote smoothness of the underlying surface on which the point clouds are discrete samples. Along with coordinates, we thus form a 6-dimensional feature vector at each point i, i.e., f i = [x i , y i , z i , n i x , n i y , n i z ] , where [x i , y i , z i ] denotes the coordinates of point i, and [n i x , n i y , n i z ] denotes its normal vector.…”

3D Point Cloud Denoising via Deep Neural Network Based Local Surface Estimation

“…1. System pipeline: During pre-processing, we calculate reference planes for points in noiseless point clouds as in [20]. During denoising, we process noisy point clouds to estimate reference planes for points.…”

“…1. During pre-processing, we use graph-based techniques to calculate reference planes for points as in [20]. Let T i be the reference plane with normal vector a i and interception c i calculated from noiseless point cloud for point p i .…”