Processing and registration of range images from time of flight laser systems

Matthew R. Kitchin, S. H. Marín, Andrew M. Wallace, Gavin J. Gibson

Research output: Contribution to journalArticle

Abstract

We report the theory and implementation of new approaches for the processing of 3D range data in pursuit of library-based object recognition and registration. The image data is obtained from an active LaDAR system (scanned Time-Correlated Single Photon Count or time-gated Burst Illumination Laser) and describes the range and 3D surface characteristics of remote objects at specific views. The reflected laser signal returns are generally embedded in noise and clutter of uncertain origin. We have applied the Markov Chain Monte Carlo (MCMC) methodology, using random sampling of the search space, to evaluate the number, positions and amplitudes of returns in such scenarios. We describe the use of methods for removing outliers and smoothing these time-of-flight generated depth images, based on least median of squares and anisotropic diffusion, respectively. Further, we outline and demonstrate procedures for registration and pose determination of objects from range data. This consists of three phases, namely point feature extraction, pose clustering and registration. The first computes a surface metric facilitating candidate correspondence determination, using the technique of pair-wise geometric histograms. The second is carried out by a leader-based algorithm, which does not require the number of clusters to be pre-specified. The third is an extension of the iterative closest points (ICP) method, being specifically designed for mesh representations. Collectively, these processes allow an object within a scene - described by a 3D range image - to be matched with a preformed model from a database.

Original languageEnglish
Article number598803
JournalProceedings of SPIE - the International Society for Optical Engineering
Volume5988
DOIs
Publication statusPublished - 2005
EventUnmanned/Unattended Sensors and Sensor Networks II - Bruges, Belgium
Duration: 26 Sep 200528 Sep 2005

Fingerprint

random sampling
lasers
Markov chains
clutter
histograms
smoothing
pattern recognition
mesh
bursts
illumination
methodology
photons

Keywords

  • Image registration
  • LaDAR
  • MCMC
  • Target recognition

Cite this

@article{09543e284e9d49af8e02633befd10e22,
title = "Processing and registration of range images from time of flight laser systems",
abstract = "We report the theory and implementation of new approaches for the processing of 3D range data in pursuit of library-based object recognition and registration. The image data is obtained from an active LaDAR system (scanned Time-Correlated Single Photon Count or time-gated Burst Illumination Laser) and describes the range and 3D surface characteristics of remote objects at specific views. The reflected laser signal returns are generally embedded in noise and clutter of uncertain origin. We have applied the Markov Chain Monte Carlo (MCMC) methodology, using random sampling of the search space, to evaluate the number, positions and amplitudes of returns in such scenarios. We describe the use of methods for removing outliers and smoothing these time-of-flight generated depth images, based on least median of squares and anisotropic diffusion, respectively. Further, we outline and demonstrate procedures for registration and pose determination of objects from range data. This consists of three phases, namely point feature extraction, pose clustering and registration. The first computes a surface metric facilitating candidate correspondence determination, using the technique of pair-wise geometric histograms. The second is carried out by a leader-based algorithm, which does not require the number of clusters to be pre-specified. The third is an extension of the iterative closest points (ICP) method, being specifically designed for mesh representations. Collectively, these processes allow an object within a scene - described by a 3D range image - to be matched with a preformed model from a database.",
keywords = "Image registration, LaDAR, MCMC, Target recognition",
author = "Kitchin, {Matthew R.} and Mar{\'i}n, {S. H.} and Wallace, {Andrew M.} and Gibson, {Gavin J.}",
year = "2005",
doi = "10.1117/12.629891",
language = "English",
volume = "5988",
journal = "Proceedings of SPIE - the International Society for Optical Engineering",
issn = "1996-756X",
publisher = "SPIE",

}

TY - JOUR

T1 - Processing and registration of range images from time of flight laser systems

AU - Kitchin, Matthew R.

AU - Marín, S. H.

AU - Wallace, Andrew M.

AU - Gibson, Gavin J.

PY - 2005

Y1 - 2005

N2 - We report the theory and implementation of new approaches for the processing of 3D range data in pursuit of library-based object recognition and registration. The image data is obtained from an active LaDAR system (scanned Time-Correlated Single Photon Count or time-gated Burst Illumination Laser) and describes the range and 3D surface characteristics of remote objects at specific views. The reflected laser signal returns are generally embedded in noise and clutter of uncertain origin. We have applied the Markov Chain Monte Carlo (MCMC) methodology, using random sampling of the search space, to evaluate the number, positions and amplitudes of returns in such scenarios. We describe the use of methods for removing outliers and smoothing these time-of-flight generated depth images, based on least median of squares and anisotropic diffusion, respectively. Further, we outline and demonstrate procedures for registration and pose determination of objects from range data. This consists of three phases, namely point feature extraction, pose clustering and registration. The first computes a surface metric facilitating candidate correspondence determination, using the technique of pair-wise geometric histograms. The second is carried out by a leader-based algorithm, which does not require the number of clusters to be pre-specified. The third is an extension of the iterative closest points (ICP) method, being specifically designed for mesh representations. Collectively, these processes allow an object within a scene - described by a 3D range image - to be matched with a preformed model from a database.

AB - We report the theory and implementation of new approaches for the processing of 3D range data in pursuit of library-based object recognition and registration. The image data is obtained from an active LaDAR system (scanned Time-Correlated Single Photon Count or time-gated Burst Illumination Laser) and describes the range and 3D surface characteristics of remote objects at specific views. The reflected laser signal returns are generally embedded in noise and clutter of uncertain origin. We have applied the Markov Chain Monte Carlo (MCMC) methodology, using random sampling of the search space, to evaluate the number, positions and amplitudes of returns in such scenarios. We describe the use of methods for removing outliers and smoothing these time-of-flight generated depth images, based on least median of squares and anisotropic diffusion, respectively. Further, we outline and demonstrate procedures for registration and pose determination of objects from range data. This consists of three phases, namely point feature extraction, pose clustering and registration. The first computes a surface metric facilitating candidate correspondence determination, using the technique of pair-wise geometric histograms. The second is carried out by a leader-based algorithm, which does not require the number of clusters to be pre-specified. The third is an extension of the iterative closest points (ICP) method, being specifically designed for mesh representations. Collectively, these processes allow an object within a scene - described by a 3D range image - to be matched with a preformed model from a database.

KW - Image registration

KW - LaDAR

KW - MCMC

KW - Target recognition

UR - http://www.scopus.com/inward/record.url?scp=33644500740&partnerID=8YFLogxK

U2 - 10.1117/12.629891

DO - 10.1117/12.629891

M3 - Article

VL - 5988

JO - Proceedings of SPIE - the International Society for Optical Engineering

JF - Proceedings of SPIE - the International Society for Optical Engineering

SN - 1996-756X

M1 - 598803

ER -