LPD-AE: Latent Space Representation of Large-Scale 3D Point Cloud

Chuanzhe Suo,Lingfei Mo,Yunhui Liu,Zhe Liu

doi:10.1109/access.2020.2999727

Abstract

The effective latent space representation of point cloud provides a foremost and fundamental manner that can be used for challenging tasks, including point cloud based place recognition and reconstruction, especially in large-scale dynamic environments. In this paper, we present a novel deep neural network, LPD-AE(Large-scale Place Description AutoEncoder Network), to obtain meaningful local and contextual features for the generation of latent space from 3D point cloud directly. The encoder network constructs the discriminative global descriptors to realize high accuracy and robust place recognition, which contributed by extracting the local neighbor geometric features and aggregating neighborhood relationships both in feature space and physical space. The decoder network performs hierarchical reconstruction on coarse key points and ultimately produce dense point clouds, which shows that it is capable of reconstructing a full point cloud frame from a single compact but high dimensional descriptor. Our proposed network demonstrates performance that is comparable to the state-of-the-art approaches. With the benefit of the LPD-AE, many computationally complex tasks that rely directly on point clouds can be effortlessly conducted on latent space with lower memory costs, such as relocalization, loop closure detection, and map compression reconstruction. Comprehensive validations on Oxford RobotCar dataset, KITTI dataset, and our freshly collected dataset, which contains multiple trials of repeated routes in different weather and at different times, manifest its potency for real robotic and self-driving implementation. The source code is available at https://github.com/Suoivy/LPD-AE .

Highlights

LIDAR significantly boosts the progress of self-driving and robotic technologies and becomes the primary sensor to sense the environment for increasing technology maturity and decreasing cost
As a pioneer of neural network feature extraction, PointNet [5] laid the foundation of deep learning on the point cloud by applying a symmetric function to each point independently
To address the above issues, we present a novel and complete latent space representation pipeline, LPD-AE (Large-scale Place Description AutoEncoder Network), which consists of the following two parts for recognition and reconstruction: FIGURE 1

Summary

Introduction

LIDAR significantly boosts the progress of self-driving and robotic technologies and becomes the primary sensor to sense the environment for increasing technology maturity and decreasing cost It can directly depict the real physical world in point clouds, containing real scale measurements and geometric features, which has natural advantages in SLAM(Simultaneous Localization And Mapping) [1]. Effective latent space representation of point cloud provides a reliable solution, which represents the point cloud by a single global feature utilized to place recognition for solving loop closure and relocalization tasks. To this end, deep learning on the 3D point cloud affords a powerful tool because of its excellent performance on feature extraction and generalization ability. The improved PointNet++ [6] and subsequent DGCNN [7] introduced neighborhood rather than individual points through hierarchical sampling and dynamic graph network, respectively, to better collect

Objectives

Methods

Findings

Conclusion