Advancements in deep learning for point cloud classification and segmentation: A comprehensive review

Abstract

Point clouds, a foundational 3D data representation, are extensively utilized in fields such as autonomous driving and robotics due to their capability to represent complex spatial structures. With the rapid advancement of artificial intelligence, leveraging deep learning to enhance point cloud processing has become a central focus in computer vision research. The unstructured nature, large-scale data volume, and labor-intensive annotation of point clouds present unique challenges for designing deep learning models. This paper provides a comprehensive review of the development and latest advancements in deep learning models for point cloud processing, with a specific focus on classification and segmentation. We systematically outline the technical approaches and key strategies for addressing these challenges, offering a clear understanding of the most recent and notable research in the field. Furthermore, we discuss the potential challenges and future research directions in point cloud processing by analyzing the respective strengths and weaknesses of prevailing techniques, thus to guide the evolution of point cloud processing technologies.