Accession Number:

AD1188444

Title:

3DLIVE Techinque Analysis: A Study of Segmentation, Classification and Object Detection of 3D Point Cloud Datasets

Descriptive Note:

[Technical Report, Technical Paper]

Corporate Author:

AFRL/RI - Information Directorate

Personal Author(s):

• Moquin, Damain
• Thorpe, Claire
• Williams, Caleb
• Schwartz, Casey
• Turk, Dakota
• Emad, Ariana

Report Date:

2022-12-20

Pagination or Media Count:

49

Abstract:

The aim of this research is to discuss the current state-of-the-art practices and methods for machine learning algorithms that perform on point cloud data. The research conducted will be applied to the in-house efforts of the Three Dimensional Lidar Visualization and Exploitation 3DLIVE team, whose primary goal is to create a new system for visualization and interaction with point cloud data for Target Coordinate Mensuration TCM. The proposed machine learning methods relate to three main topics in machine learning for 3D point clouds and computer vision, each of which had its own segment of papers researched.These topics are segmentation, classification and object detection, and the selected papers are of recent studies that achieved state-of-the-art performances. The findings of this research are a select few methods that show the most promising results and effectiveness to the 3DLIVE team. Effectiveness is largely dependent on the scalability and applicability of the algorithm to the3DLIVE use case as well as its accuracy and precision.

Descriptors:

• machine learning
• artificial intelligence
• target recognition
• computer vision
• computational science
• information science
• computers
• deep learning
• detectors
• three dimensional
• automated target recognition
• detection
• neural networks
• artificial intelligence software
• convolutional neural networks
• urban areas
• accuracy

Subject Categories:

Distribution Statement:

[A, Approved For Public Release]