An Incremental Learning Method for Automatic Annotation of Point Cloud Data

Read full text Summarise with VIRAT
Authors: XIE Zhe, Zhang Jiawei, LONG Teng, LI Peijie
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

With the widespread application of intelligent systems in autonomous driving, industrial automation and more fields, the industrial requirements for machines’ environmental perception capability have risen. As a core data form for 3D perception, point clouds play a crucial role in environment mapping, object detection, and path planning. However, the complex structure and large volume of point cloud data make traditional manual annotation inefficient, costly, and error-prone, and thus inadequate for large-scale intelligent perception systems. This paper proposes an automatic annotation method for point cloud data that integrates the CenterPoint 3D detection model with an incremental learning mechanism. The CenterPoint model efficiently extracts spatial features and performs object detection, while the incremental learning strategy based on an “inference-correction-feedback-optimization” process progressively improves model accuracy and generalization. Experimental results show that this method requires only 20% of the initial manually annotated data and achieves a single-frame inference time of merely 0.06 seconds, with reduced manual intervention costs by more than 62% compared with purely manual annotation. By improving annotation efficiency and consistency of point cloud data, the proposed method provide high-quality training data support for intelligent perception systems, and thus has promising research value and practical application prospects.

Fields of Study & Subjects

Advanced Neural Network Applications 3D Shape Modeling and Analysis Domain Adaptation and Few-Shot Learning

Publication details

Journal / source
DOAJ (DOAJ: Directory of Open Access Journals)
Publication date
1 February 2026
Language
ZH