Deep Learning for 3D Point Cloud Understanding: A Survey

Haoming Lu,Humphrey Shi

2021-05-23

Abstract:The development of practical applications, such as autonomous driving and robotics, has brought increasing attention to 3D point cloud understanding. While deep learning has achieved remarkable success on image-based tasks, there are many unique challenges faced by deep neural networks in processing massive, unstructured and noisy 3D points. To demonstrate the latest progress of deep learning for 3D point cloud understanding, this paper summarizes recent remarkable research contributions in this area from several different directions (classification, segmentation, detection, tracking, flow estimation, registration, augmentation and completion), together with commonly used datasets, metrics and state-of-the-art performances. More information regarding this survey can be found at: <a class="link-external link-https" href="https://github.com/SHI-Labs/3D-Point-Cloud-Learning" rel="external noopener nofollow">this https URL</a>.

Computer Vision and Pattern Recognition,Machine Learning

What problem does this paper attempt to address?

The paper primarily addresses several key issues in 3D point cloud understanding, especially in the context of deep learning frameworks. Specifically:

1. **High Information Density Representation**: How to find high information density representation methods from sparse point clouds.
2. **Network Design**: How to construct neural networks that meet specific constraints (such as size invariance and permutation invariance).
3. **Efficient Processing of Large Data Volumes**: How to process large amounts of data with low time and computational resource consumption.

To showcase the latest advancements in these areas, the paper summarizes recent research achievements in classification, segmentation, detection, tracking, flow estimation, registration, enhancement, and completion. It also introduces commonly used databases, evaluation metrics, and current best performances. Additionally, the paper discusses various methods, including projection-based methods (such as multi-view representation and voxel representation) and point-based methods (such as using multi-layer perceptrons and graph networks), providing detailed introductions and comparisons of these methods. Through this work, the paper aims to advance the development of 3D point cloud understanding technology and provide better technical support for practical applications such as autonomous driving and robotics.