DGCB-Net: Dynamic Graph Convolutional Broad Network for 3D Object Recognition in Point Cloud

Yunsick Sung,Yifei Tian,Sangchul Woo,Wei Song,Long Chen

doi:10.3390/rs13010066

Yunsick Sung, Yifei Tian + Show 3 more

Open Access

https://doi.org/10.3390/rs13010066

Copy DOI

Abstract

3D (3-Dimensional) object recognition is a hot research topic that benefits environment perception, disease diagnosis, and the mobile robot industry. Point clouds collected by range sensors are a popular data structure to represent a 3D object model. This paper proposed a 3D object recognition method named Dynamic Graph Convolutional Broad Network (DGCB-Net) to realize feature extraction and 3D object recognition from the point cloud. DGCB-Net adopts edge convolutional layers constructed by weight-shared multiple-layer perceptrons (MLPs) to extract local features from the point cloud graph structure automatically. Features obtained from all edge convolutional layers are concatenated together to form a feature aggregation. Unlike stacking many layers in-depth, our DGCB-Net employs a broad architecture to extend point cloud feature aggregation flatly. The broad architecture is structured utilizing a flat combining architecture with multiple feature layers and enhancement layers. Both feature layers and enhancement layers concatenate together to further enrich the features’ information of the point cloud. All features work on the object recognition results thus that our DGCB-Net show better recognition performance than other 3D object recognition algorithms on ModelNet10/40 and our scanning point cloud dataset.

Highlights

The significant advantage of 3D point clouds is its numerous original geometric and topology information, the disadvantage is that its inherent characteristics always cause a series of bottleneck problems in feature extraction [3]
Inspired by the dynamic graph convolutional neural networks (DGCNNs) in the deep learning domain [9], this paper developed a dynamic graph convolutional broad network (DGCB-Net) that maps and flatly extends graph features to enrich point cloud features
This paper proposed a DGCB-Net model to realize object recognition from point clouds by combining convolution and broad learning systems

Summary

Introduction

The significant advantage of 3D point clouds is its numerous original geometric and topology information, the disadvantage is that its inherent characteristics always cause a series of bottleneck problems in feature extraction [3]. To avoid the influence caused by these characteristics, point cloud pre-processes, or registration steps are required in the majority of 3D object recognition methods [5]. For sampling uniform features from 3D object models, some researchers use volumetric models with a pre-defined resolution to resample and initial the point cloud [6]. Multiview transform is another common method to transform the uncertain geometric structure into fixed temporal or frequency domains [7]. These projection-based feature representation models contain a certain degree of information loss during transforming processes. Some local feature grouping methods (e.g., farthest point sampling, k-nearest neighboring) are adopted to extract local point cloud features

Methods

Findings

Discussion

Conclusion