Sequential Point Cloud Prediction in Interactive Scenarios: A Survey

TL;DR

This survey reviews methods for sequential point cloud prediction in interactive scenarios, emphasizing frameworks, feature extraction, and tasks like scene flow estimation and point cloud location prediction, highlighting future research directions.

Contribution

It provides a comprehensive overview of PCS prediction methods, defining the problem, analyzing feature extraction, and comparing key tasks in interactive scenarios.

Findings

Summarizes existing PCS prediction frameworks

Analyzes spatio-temporal feature extraction methods

Highlights differences between scene flow and location prediction

Abstract

Point cloud has been widely used in the field of autonomous driving since it can provide a more comprehensive three-dimensional representation of the environment than 2D images. Point-wise prediction based on point cloud sequence (PCS) is an essential part of environment understanding, which can assist in the decision-making and motion-planning of autonomous vehicles. However, PCS prediction has not been deeply researched in the literature. This paper proposes a brief review of the sequential point cloud prediction methods, focusing on interactive scenarios. Firstly, we define the PCS prediction problem and introduce commonly-used frameworks. Secondly, by reviewing non-predictive problems, we analyze and summarize the spatio-temporal feature extraction methods based on PCS. On this basis, we review two types of PCS prediction tasks, scene flow estimation (SFE) and point cloud location prediction (PCLP), highlighting their connections and differences. Finally, we discuss some opening issues and point out some potential research directions.