Recent Advances and Perspectives in Deep Learning Techniques for 3D Point Cloud Data Processing

Abstract

In recent years, deep learning techniques for processing 3D point cloud data have seen significant advancements, given their unique ability to extract relevant features and handle unstructured data. These techniques find wide-ranging applications in fields like robotics, autonomous vehicles, and various other computer-vision applications. This paper reviews the recent literature on key tasks, including 3D object classification, tracking, pose estimation, segmentation, and point cloud completion. The review discusses the historical development of these methods, explores different model architectures, learning algorithms, and training datasets, and provides a comprehensive summary of the state-of-the-art in this domain. The paper presents a critical evaluation of the current limitations and challenges in the field, and identifies potential areas for future research. Furthermore, the emergence of transformative methodologies like PoinTr and SnowflakeNet is examined, highlighting their contributions and potential impact on the field. The potential cross-disciplinary applications of these techniques are also discussed, underscoring the broad scope and impact of these developments. This review fills a knowledge gap by offering a focused and comprehensive synthesis of recent research on deep learning techniques for 3D point cloud data processing, thereby serving as a useful resource for both novice and experienced researchers in the field.

FusionVision: A Comprehensive Approach of 3D Object Reconstruction and Segmentation from RGB-D Cameras Using YOLO and Fast Segment Anything

“…These images serve as a baseline for assessing the model's performance in familiar settings and providing insights into its ability to handle variations within its training domain. (2) The second set of images introduces variability in object positions, orientations, and lighting conditions compared to the training data. By capturing a broader range of scenarios, this set enables to evaluate the model's adaptability to changes in object positions, orientations, and lighting, while simulating real-world challenges such as occlusions and shadows.…”

“…The significance of point-cloud processing has surged across various domains, such as robotics [1,2], medical field [3,4], autonomous driving [5,6], metrology [7][8][9], etc. Over the past few years, advancements in vision sensors have led to remarkable improvements, enabling these sensors to provide real-time 3D measurements of the surroundings while maintaining decent accuracy [10,11].…”

A Survey of 6DoF Object Pose Estimation Methods for Different Application Scenarios

“…In contrast, methods based on point clouds or depth maps may offer unexpected advantages, while RGB images lack geometric data. Depth information or point cloud information contains rich shape geometry information, which is significant for inferring the pose of objects [ 83 , 84 ].…”

Practicalities of Incorporating 3D Laser Scanning with BIM in Live Construction Projects: A Case Study

“…Table 2 illustrates some of the relevant specs for the computer used for this project. The size of the point cloud relies on various factors, such as the point density, the duration and frequency of the data acquisition, and number of scans required when capturing data [24,53]. The size of the point clouds captured from the focus project significantly increased due to the size and progress of the construction project, the site occlusions, and an increase in scan numbers, which directly impacted the data acquisition, pre-processing, integration, and alignment strategies in this project.…”