Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Surface Classification with Millimeter-Wave Radar for Constant Velocity Devices using Temporal Features and Machine Learning

Montgomery, David LU and Holmén, Gaston LU (2019) In Master's Theses in Mathematical Sciences FMSM01 20182
Mathematical Statistics
Abstract
Classification of surfaces in the near field using millimeter-wave radar commonly considers the use of polarization based methods for road condition monitoring. When a surface consists of larger structures one instead wishes to monitor the surface topography. Analysis of scattering from rough surfaces is highly complex and relies on prior knowledge of surface structure. In this work a device moving at constant velocity is considered. By constructing a set of slow and fast time based features a machine learning classifier is used to distinguish grass target surfaces from asphalt, gravel, soil and tiled surfaces. It is found that using estimated autocovariances and average envelope shapes make for efficient features and that a small fully... (More)
Classification of surfaces in the near field using millimeter-wave radar commonly considers the use of polarization based methods for road condition monitoring. When a surface consists of larger structures one instead wishes to monitor the surface topography. Analysis of scattering from rough surfaces is highly complex and relies on prior knowledge of surface structure. In this work a device moving at constant velocity is considered. By constructing a set of slow and fast time based features a machine learning classifier is used to distinguish grass target surfaces from asphalt, gravel, soil and tiled surfaces. It is found that using estimated autocovariances and average envelope shapes make for efficient features and that a small fully connected neural network classifier adequately manages to determine the surface type. The found model is accurate yet parsimonious and could be implemented with limited hardware requirements. Application of a median filter onto the sequence of classifier predictions effectively suppresses outlying predictions. This model can find use in autonomous devices that have tasks performed on designated surface types, such as in autonomous lawn mowers. (Less)
Popular Abstract
In this work we have tested the use of a small radar sensor for surface classification for autonomous robots. Such a system could be useful for robots operating on a particular surface type. The experiments presented in this report have focused on distinguishing grass surfaces from other types of surfaces. By the use of mathematical models we found that such separation is possible.
Please use this url to cite or link to this publication:
author
Montgomery, David LU and Holmén, Gaston LU
supervisor
organization
course
FMSM01 20182
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Machine Learning, Surface Classification, Feature Extraction, Autonomous Devices, Millimeter-wave Radar
publication/series
Master's Theses in Mathematical Sciences
report number
LUTFMS-3361-2019
ISSN
1404-6342
other publication id
2019:E4
language
English
id
8974567
date added to LUP
2019-04-25 11:02:01
date last changed
2024-09-30 11:44:10
@misc{8974567,
  abstract     = {{Classification of surfaces in the near field using millimeter-wave radar commonly considers the use of polarization based methods for road condition monitoring. When a surface consists of larger structures one instead wishes to monitor the surface topography. Analysis of scattering from rough surfaces is highly complex and relies on prior knowledge of surface structure. In this work a device moving at constant velocity is considered. By constructing a set of slow and fast time based features a machine learning classifier is used to distinguish grass target surfaces from asphalt, gravel, soil and tiled surfaces. It is found that using estimated autocovariances and average envelope shapes make for efficient features and that a small fully connected neural network classifier adequately manages to determine the surface type. The found model is accurate yet parsimonious and could be implemented with limited hardware requirements. Application of a median filter onto the sequence of classifier predictions effectively suppresses outlying predictions. This model can find use in autonomous devices that have tasks performed on designated surface types, such as in autonomous lawn mowers.}},
  author       = {{Montgomery, David and Holmén, Gaston}},
  issn         = {{1404-6342}},
  language     = {{eng}},
  note         = {{Student Paper}},
  series       = {{Master's Theses in Mathematical Sciences}},
  title        = {{Surface Classification with Millimeter-Wave Radar for Constant Velocity Devices using Temporal Features and Machine Learning}},
  year         = {{2019}},
}