Please use this identifier to cite or link to this item: https://hdl.handle.net/10356/157862
Title: Object detection using OTC LiDAR sensors
Authors: Tan, Mark Jen Wei
Keywords: Engineering::Electrical and electronic engineering::Computer hardware, software and systems
Issue Date: 2022
Publisher: Nanyang Technological University
Source: Tan, M. J. W. (2022). Object detection using OTC LiDAR sensors. Final Year Project (FYP), Nanyang Technological University, Singapore. https://hdl.handle.net/10356/157862
Project: A3284-211
Abstract: Autonomous navigation of unmanned aerial vehicles (UAVs) relies heavily on the capabilities of both onboard and attached visual sensors to provide data that can be processed for intelligent decision making during aerial operations. In this paper, we consider the problem of existing navigational tools such as RGB cameras and Global Positioning Systems (GPS) to be limited in functionality for certain usecases within autonomous navigation. Recent literature has suggested that LiDAR sensors show significant potential in adding value to this field due to their ability to transmit their own signals and create precise 3D coordinate data. However, most commercially in-use LiDAR sensors are extremely expensive, making it economically unviable to conduct extensive testing. In this work, we propose the use of low-cost off-the-counter (OTC) LiDAR sensors to conduct object detection as a proof of concept for their use in autonomous navigations use-cases. The usage of these sensors will enable us to mitigate the financial constraints of extensive testing. We also propose the use of a deep learning point cloud object detection model, PointPillars, as a complementing method for our OTC LiDAR sensor due to the network’s ability to have a balance of low computational requirements, fast speeds and high accuracy when compared to similar 3D object detection networks. For this project, we ran tests involving data collected from the L515 to detect vehicular objects using a pre-trained PointPillars network. Extensive testingshows that despite inaccuracies involving our deep learning model detecting objects from data collected using the L515, our concept has been proven with moderate success. We inferred based on our results that low-cost LiDAR sensors could add value to indoor autonomous navigation, as well as use cases in environments without significant ambient light and where range is not a demanding factor. Furthermore, a pipeline, accompanying functions and a GUI for the L515 on MATLAB’s platform has been shared to provide future researchers with the tools to conduct more tests in this area. Finally, we have documented several key issues with respect to the L515, as well as possible solutions that can be explored in future work. This information will prove useful when extrapolated to other short range LiDAR sensors
URI: https://hdl.handle.net/10356/157862
Fulltext Permission: restricted
Fulltext Availability: With Fulltext
Appears in Collections:EEE Student Reports (FYP/IA/PA/PI)

Files in This Item:
File Description SizeFormat 
FYP Revised Final Report (1).pdf
  Restricted Access
2.55 MBAdobe PDFView/Open

Page view(s)

29
Updated on Dec 9, 2022

Download(s)

1
Updated on Dec 9, 2022

Google ScholarTM

Check

Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.