PCEDNet: A Lightweight Neural Network for Fast and Interactive Edge Detection in 3D Point Clouds

Abstract

In recent years, Convolutional Neural Networks (CNN) have proven to be efficient analysis tools for processing point clouds, e.g., for reconstruction, segmentation, and classification. In this article, we focus on the classification of edges in point clouds, where both edges and their surrounding are described. We propose a new parameterization adding to each point a set of differential information on its surrounding shape reconstructed at different scales. These parameters, stored in a Scale-Space Matrix (SSM) , provide a well-suited information from which an adequate neural network can learn the description of edges and use it to efficiently detect them in acquired point clouds. After successfully applying a multi-scale CNN on SSMs for the efficient classification of edges and their neighborhood, we propose a new lightweight neural network architecture outperforming the CNN in learning time, processing time, and classification capabilities. Our architecture is compact, requires small learning sets, is very fast to train, and classifies millions of points in seconds.

Earthwork digital twin for teleoperation of an automated bulldozer in edge dumping

“…Different 3D model detection algorithms are used for comparative analysis to verify the advancement of the retaining wall edge detection algorithm proposed in this study. Figure 18 shows the comparison between several other recent related works for point cloud edge detection: FEE (Mérigot et al, 2010), PCEDNet‐2c (Himeur et al, 2021), ECNet (Yu et al, 2018), and the model proposed in this study. Precision and F 1 are two common indexes used to evaluate edge detection results.…”

ImmersiveIML – Immersive interactive machine learning for 3D point cloud classification: the neural network at your fingertips

“…To achieve a high-quality dialogue we require to use a very fast model, both for training and inference. We chose PCEDNet [13], which provides both training on thousands of points and classification of million of points in seconds. This model is thus a lightweight neural network fast enough for supporting an interactive human/system dialogue.…”

SGLBP: Subgraph‐based Local Binary Patterns for Feature Extraction on Point Clouds

“…Mostly recently, Himeur et al . [HLP*21] designed a network by proposing a new parameterization adding to each point a set of differential information on its surrounding shape reconstructed at different scales; and Bazazian et al . proposed [BP21] EDCNet, which uses both edge labels and segmentation labels as the Ground Truth.…”

“…More precisely, we also utilize two other metrics presented in [HLP*21] to further evaluate and compare these methods, i.e. the matthews correlation coefficient ($$MCC$$) (Equation (9)) and the Intersection over union score ($$IoU$$)) (Equation (10)), which are calculated based on the values of true positives (TP), false positives (FP), true negatives (TN) and false negatives (FN).…”