Sentinel Prime--A Novel Approach to 3D Classification

Mohamed, Harris

Permalink

https://hdl.handle.net/2142/110309 Copy

Description

Title

Sentinel Prime--A Novel Approach to 3D Classification

Author(s)

Mohamed, Harris

Contributor(s)

Deming, Chen

Issue Date

2021-05

Keyword(s)

• robotics
• SLAM
• classification
• Lidar
• sensor fusion

Abstract

3D object detection and classification from point cloud data and monocular cameras is an essential task for many fields, such as autonomous navigation and augmented reality applications. Existing solutions that operate solely on the point cloud data will represent the data in a sparse manner to then input into a convolutional neural net. However, this approach tends to be inaccurate as the transformation of the data will always lose resolution. Solutions that operate on RGB images of the environment perform quite well, as these have been around for several years. The purpose of Sentinel Prime is to develop a robot to run a sensor fusion network which will combine 2D image data and 3D LIDAR data to beat the performance of a solely 2D or 3D network. The completed robot will be demoed on an indoor environment, shown to be correctly classifying several indoor objects.

Type of Resource

text

Language

en

Permalink

http://hdl.handle.net/2142/110309

Owning Collections