Research progress of part point cloud detection based on deep learning

As the demand for 3D vision technology in industry continues to rise, 3D point cloud part classification based on deep learning has received widespread attention. Classification methods based on 3D point cloud data outperform traditional 2D image classification in terms of inspection accuracy, and are advantageous in detecting parts with similar and poorly differentiated part shapes and parts with uncertainty in the type and extent of damage areas. The aim of this paper is to introduce some of the mainstream part inspection methods based on 3D point cloud data. Firstly, the advantages and background of 3D point clouds in industrial applications and the characteristics of their acquisition methods are introduced; secondly, the classification methods based on 3D point clouds of parts by classical and deep learning algorithms and the related industrial applications are introduced; the common datasets for 3D point cloud classification are summarised; Finally, the development trends and potential research directions of deep learning in the direction of part point cloud detection are foreseen.