Automatic measurement of difficult solid objects in unstructured and challenging environments
Submitting Institution
University of the West of England, BristolUnit of Assessment
General EngineeringSummary Impact Type
TechnologicalResearch Subject Area(s)
Information and Computing Sciences: Artificial Intelligence and Image Processing
Medical and Health Sciences: Neurosciences
Summary of the impact
Companies in the manufacturing, construction, courier, freight and
airline industries have made use of Photometric Stereo based imaging
technology, developed by the Centre for Machine Vision (CMV) at UWE
Bristol, to capture 2D and 3D data simultaneously. This has enabled them
to develop new products for large volumetric measurement, characterisation
of aggregates, tile quality control and automotive wheel alignment. These
examples have addressed hitherto challenging tasks or have extended
functionality to new or poorly structured environments, for instance: (i)
the robust capture of accurate 3D data from postal packages exhibiting
complex shape and coloration; and (ii) a new capability for distinction
between printing and moulding defects during fast moving tile quality
control — previously impossible. In all cases ultra-low-cost equipment is
used — underlining the attractiveness of the techniques developed by the
CMV.
Underpinning research
Research at UWE's Centre for Machine Vision (CMV) has developed
photometric stereo (PS) imaging techniques that can be used in
unstructured, real-world environments, away from the ordered and
controlled environment of the laboratory. Three key advances of the
underpinning research are as follows:
- CMV has developed techniques to extract a "bump map" from objects — a
3D texture that can be "peeled off" a surface, allowing the colour, 3D
texture and underlying surface to be isolated and inspected from any
angle;
- PS techniques from elsewhere can only image in a static way, whereas
CMV's approach was the first to succeed in doing this for a moving
object or one with non-ridged surfaces; and
- others have worked in a lab setting with idealised illumination, i.e.
a small specialist lighting source at a distance, whereas CMV's method
has worked with cheap off-the-shelf lights placed close by.
The following people were involved: Prof Melvyn Smith (Director
of the CMV); Dr Lyndon Smith (Co-director of the CMV); Dr Gary
Atkinson (Senior Lecturer, CMV); Dr Abdul Farooq (Senior
Lecturer, CMV); and Dr Jiuai Sun (Lecturer, CMV).
Photometric stereo (PS) for practical surface metrology, inspection and
analysis was pioneered at UWE's CMV (formerly named the Machine Vision
Laboratory) in the 1990s (R1). Before this, no practical means existed to
capture detailed 3D topography at sub-pixel resolution to identify
objects, textures or defects while camouflaged by complex surface
reflectance/colour patterns, which obscure topography.
Prior work had tended to concentrate on the `reverse engineering' of 3D
objects, for which the application of conventional PS per se was
problematic. Early research insight at CMV by M. Smith in 1999
identified PS as advantageous for the capture of fine 3D surface features
at sub-pixel resolution from predominantly planar surfaces. A key
development by M. Smith, L. Smith and Farooq in
2005, Dynamic PS, served to extend the technique from static to moving
scenes. M. Smith and Farooq established techniques from
2000 to 2006 that could isolate concomitant 3D and 2D surface features for
inspection of fast moving decorative ceramic tiles (R2) and other natural
materials, such as polished stone (R3).
In 2003 M. Smith and L. Smith also identified PS as
offering a useful synergy with other established 3D imaging techniques,
such as laser and stereo triangulation, where novel hybrids offered new
functionality by combining the excellent high-resolution but poor
low-resolution performance of PS with the complementary attributes of
conventional 3D imaging. In addition, because PS allowed the appearance of
surface colour to be depressed, while isolating and amplifying 3D shape
attributes, the technique was identified as potentially useful in
simplifying a wide range of conventionally difficult unstructured imaging
tasks.
CMV was able to overcome a major obstacle to the practical application of
PS: the step from a static to moving imaging technology (M. Smith
in 2005), together with a number of other limiting practical issues. The
solution was realised by allowing PS to operate in unconstrained moving,
cluttered, changing or confined environments and to integrate the
technology with existing techniques. This was first achieved by M.
Smith, L. Smith and Farooq between 2000 and 2005 by
introducing a patented multiplexing approach — either space, frequency or
temporally based — to realise a new dynamic form of PS.
By 2005, the investigation of novel hybrid PS methods and advanced
modelling by M. Smith and L. Smith culminated in a
practical dynamic or moving PS demonstrator (R5). In 2007, this research
established the minimum number of illuminates (six) to allow any convex
object to be fully recovered, and later (by Sun and M. Smith
in 2007) to new models allowing use of practical low-cost hardware, such
as non-specialist illumination (R4). These findings were particularly
important in widening application and in realising practical commercial
applications of PS able to address conventionally challenging imaging
tasks, such as the segmentation, characterisation and metrology of
embedded 3D objects in complex scenes and in using PS at close range (1mm)
or at a large stand-off (50m and more).
More recently, the outcomes of this earlier research in using PS to
robustly capture 3D data in industrial application for metrology and
complex shape characterisation have opened up new avenues at CMV exploring
further advanced forms of PS and analysis methods in unstructured
environments. These have been investigated in applications such as: (i) 3D
facial recognition by M. Smith and Atkinson (2007 to
date), where PS-derived surface normals have been shown to offer the most
effective face biometric data; (ii) skin analysis and respiratory
assessment by L. Smith (2008 to date); and (iii) facial expression
and emotion analysis (M. Smith in 2012-13), where high- resolution
3D micro-expressions have been captured for the first time.
References to the research
Publications
R1 Smith, M. L. (2001). Surface Inspection Techniques — Using the
integration of innovative machine vision and graphical modelling
techniques. John Wiley & Sons Ltd, 256 pages. ISBN 1-86058-
292-3.
R4 Sun, J., Smith, M. L., Smith, L. N., Midha, S. and Bamber, J. (2007).
Object surface recovery using a multi-light photometric stereo technique
for non-Lambertian surfaces subject to shadows and specularities. Image
and Vision Computing, 25 (7) — Special Issue on Computer Vision
Applications, 1050-1057. http://dx.doi.org/10.1016/j.imavis.2006.04.025
Grants (reverse chronological order and showing total grant
values)
Shape and outline amplification using 2.5D and 3D imaging. A new tool
for covert surveillance aimed at real threat identification such as left
object detection, M. Smith, DSTL, 2012, £60k
Intelligent Advertising Proof of Concept, M. Smith, TSB Smart
award, 2012-13, £100k
Research and develop new 3D metrology equipment within the automotive
sector, commercial, M Smith, 2011-12, £100k
Intelligent Video Surveillance KTP, A Farooq, 2010-13, £215k
Skin Reflectance and Face Shape Estimation Using Photometric Stereo
(PhotoSkin), G Atkinson, EPSRC, 2010-13, £127k
Novel non-invasive assessment of respiratory function (NORM), L
Smith, NIHR i4i Stream 1, 2008-10, £125k
Use of 3D Facial Asymmetry in Better Diagnosis and Treatment of
Plagiocephaly, M Smith, MRC, 2008-9, £120k
Innovative Technology for the Quality Control of Specular Ceramic
Materials using a Specular Photometric Stereo Sensor, M Smith, Great
Western Research and Surface Inspection Ltd, 2007-10, £75k
Face Recognition using photometric Stereo (PhotoFace), M Smith,
EPSRC, 2007-10, £672k
Application of Photometric Stereo in Dermatology (PhotoDerm), M
Smith, 2007-10, DTI, £647k
Stealthy Object Detection and Recognition, M Smith, 2007-8, MOD,
£154k
Visual inspection of polished stone, M Smith, EU 5th Framework,
2003-5, €1.3 million
Petroscope: visual 3D characterisation of aggregate materials, M
Smith, DTI Eureka Initiative, 2002-4, €1.43m
Photometric stereo for tile inspection, M Smith, EPSRC and Surface
Inspection Ltd, 2000-4, £169k
Irregular object metrology, M Smith, Commercial, multiple projects
2000-13, £119k
Details of the impact
The research developments that took place at CMV attracted much interest
from companies in the manufacturing and construction sector. This led to a
range of commercial applications in: (i) ceramic tile manufacture
inspection; (ii) polished stone defect inspection; and (iii) 3D
characterisation of aggregate materials. All of these followed from CMV's
work on evolving static PS to dynamic PS. Further application has been in:
(iv) 3D automotive wheel alignment; and (v) the segmentation and
characterisation of large objects. Both these followed from CMV work in
isolating 2D and 3D data, as detailed below.
One of the companies that approached the CMV was Quantronix Inc,
an American company operating globally, specialising in object metrology
for freight forwarding. In 2000 the company funded the CMV to develop an
Overhead Dimensioning System (ODS — see right) able to rapidly acquire 3D
data from any object(s) of up to 3m x 3m in size. CMV researchers
developed the system using the PS method (hybrid) and in 2001 the
innovation was protected and a patent application filed by the company. In
2004, a full prototype ODS was realised with a commercial product, known
as the CubiScan 1000-VS, becoming available in 2009.
First sales were made to an international freight forwarder and, since
then, over 40 systems have been sold in the US and in other countries,
including ten to FedEx and one to our own MOD at $50k each, with a total
value of over $1m. A range of benefits were delivered to their customers,
through new capability and more efficient processes. For instance,
courier, freight handlers and airlines customers, including FedEx, have
used it to measure the sizes of packages automatically, enabling them to
achieve efficiency gains by reducing both time for measurement and
previous losses due to inaccurate manual measurement, offering an average
3-week payback (S1) and other benefits such as reduced fuel consumption,
pollution and packaging materials.
Government facilities (MOD) and retail customers have been able to
optimise the use of storage space, handling assets and processes, with one
customer reporting that "one of the CubiScan 1000-VS machines they
purchased... generated over $1 million additional revenue for the
company in one year" (S1). The impact of the ODS has been continuous
since its availability and has resulted in broadening Quantronix's
in-house knowledge base to include vision-based technology. The company
has now produced a business case for developing a countertop device based
on CMV PS vision technology (S1). For this, they returned to CMV in
2012/13 to undertake the necessary R&D incorporating some of the
latest hybrid PS approaches and a new system developed by CMV is now
scheduled for release this year.
CMV's leading research in 3D metrology attracted a new collaboration from
a company in India in 2011 — Precision Testing Machines Pvt Ltd
(PTM) — which operates in the automotive sector and is the lead supplier
of sophisticated automotive diagnostic and body shop equipment
manufactured in India. A non-contact optical measuring device for
automatic automotive wheel alignment was developed by CMV, minimising
subjectivity, operator error and allowing faster throughput. In 2012, the
demonstrator was handed over to PTM engineers to develop the system to
commercial requirements. [text removed for publication]
Another application has been that of the Dynamic PS prototype developed
by CMV in 2005 for Surface Inspection Ltd, a global company which
specialises in machine vision systems and automation equipment for the
ceramic tile industry. The system developed was used to detect, isolate
and characterise concomitant 2D printing (including colour and glossiness)
and 3D moulding defects on complex surfaces with deeply relieved
topographic features and is currently in use for quality control in
ceramic tile manufacturing worldwide (S3). [text removed for publication]
The final application has been the Petroscope system developed by CMV for
Petromodel, a knowledge-based company based in Iceland, providing
software and high-technology testing instruments for quality and process
control in the aggregates industry. The Petroscope replaces labour
intensive techniques and is shaping the inspection methodology used by the
construction mineral industry (S4). A patent for this work was filed in
2004 and granted in USA and Eurasia in 2012 and in China in 2013. This
system has been demonstrated in a commercial environment for petrographic
analysis, including the characterisation of the complex 3D morphology of
aggregate particles using techniques developed as part of CMV's earlier
work in 3D face recognition, with use in the construction sector and
benefiting companies producing gravel and crushed rock for asphalt,
concrete, and railways. The system is currently being marketed and a new
version including composition (PETROSCOPE 4D) has been under test by the
Austrian railways (Jan — Aug 2013) (S4).
Sources to corroborate the impact
Testimonials listed below are available from UWE, Bristol, and
numbered on REF system in same order as below.
S1 Testimonial from CEO of Quantronix Inc, USA. Large object metrology —
corroborates customers' benefits and device sales to date, including use
by courier, freight handlers and airlines
S2 Testimonial from Managing Director of Precision Testing Machines Pvt
Ltd, India. Corroborates benefits and sales of device developed by CMV,
UWE, that automatically measures the alignment of car wheels.
S3 Testimonial from Chief Technical Officer of Surface Inspection Ltd,
UK. Corroborates benefits in surface analysis for tile inspection.
S4 Testimonial from Founder and Managing Director of Petromodel, Iceland.
Morphology of particles in petrographic analysis — corroborates benefit in
new technology for petrographic analysis.