BMVA
British Machine Vision Association and Society for Pattern Recognition
Vision in the Built Environment
One Day BMVA Technical Meeting held on 18 Feb 1998
at British Institute of Radiology, 36 Portland Place, London.
Chairpersons: Tim Ellis (City University), Gordon Mair (Strathclyde University)
10:30 Registration and coffee
10:55 Introduction, Tim Ellis (City University)
13:00 Lunch
13:55 Introduction, Gordon Mair (Strathclyde University)
14:00 3D Reconstruction of Interiors Using Laser Rangefinder and Video, Kia Ng (Leeds University)
14:30 Mobile Telepresence at Building Sites, Gordon Mair (Strathclyde University)
15:00 Tea
15:30 The Role of Vision in Construction Repair, Denis Chamberlain (City University)
16:00 Augmented Reality for Collaborative Construction, David Leevers (BICC Plc)
16:30 Closing remarks and finish
Interpreting Sequences of Images of Small-Bore
Sewers
Tony P Pridmore and Shan Fu
Department of Manufacturing Engineering and Operations Management
University of Nottingham
University Park
Nottingham
NG7 2RD
Although the construction of the UK's sewerage system was a major engineering achievement,
questionable maintenance provision has left a legacy of disrepair. It has been suggested
that around 10% of the
country's 230,000 km of sewer pipe will require reinstatement in the next decade.
Approximately 95% of the pipes concerned are classified as non-manentry; any repairs
therefore either require temporary surface works
(an expensive and disruptive process) or the deployment of appropriate trenchless
technologies. The first stage in any sewer maintenance operation is to survey the pipe. A
CCTV camera mounted on a sled or
tractor is dragged or driven through the pipe and the resulting image sequence viewed by
an operator who makes largely subjective judgements regarding the state of the sewer.
We describe the results of an EPSRC-funded project aimed at the automation of this image
interpretation process. Examination of the available data revealed a number of stable
large-scale features (the vanishing point,
displaced junctions and lateral intersections) within an otherwise largely featureless
image. An optic flow method was developed which takes into account specific properties of
sewer survey image sequences, including
concurrent motion of the camera and light source. Good quality flow fields have been
obtained and used to produce depth maps reflecting the 3D shape of the surveyed pipe.
The creation, management and exploitation of
massive image databases of process plant facilities.
Dave Chapman
University College London
Sophisticated three-dimensional CAD modelling systems are now routinely employed to ensure
cost effective design and construction of large process facilities. These systems enable
the integration of heterogeneous data
relating to the various engineering disciplines and provide new opportunities for
collaboration between designers and fabricators.
Unfortunately few design databases have been maintained beyond commissioning and
archives maintained by designers are only rarely upgraded to As-Built status. This limits
their potential for application in plant operation and, particularly, refurbishment. Thus
whilst the notion of Plant Data for Life has been widely promoted the costs of
providing complete documentation of a plants operational status has been one factor which
has inhibited the widespread promotion of such a strategy.
Recent developments in digital photogrammetry have demonstrated the capabilities of such
systems for accurate documentation of very complex facilities. However the extraction of
dimensional data from
photogrammetric archives remains an expensive and labour intensive activity and it is
seldom cost effective to convert all the information contained in the imagery to a CAD
representation.
This paper will outline a strategy that integrates the image archive with existing CAD
databases and permits ad-hoc upgrade of the model as, and when, operational requirements
indicate a positive cost benefit.
3D Reconstruction of Interiors using Laser Range-finder and Video
Dr Kia Ng, Prof. David C Hogg
Dept. Computer Science,
University of Leeds
With the advances of computational capabilities, realistic 3D models are increasingly
important to many computer applications, particularly in architecture, design, simulators
and entertainment applications.
This talk will present an integrated approach to the construction of textured 3D scene
models from laser range data and digital image, developed by an EU-ACTS project RESOLV.
The approach has been realised
within a prototype system known as the AEST (Autonomous Environmental Sensor for
Telepresence), which has been designed to fully automate the model construction of
buildings interiors, by navigating automatically
between locations at which range data and video images are captured. The principal novelty
of the project is in devising an automatic procedure for planning successive optimal
positions of the robot and integrating the
data acquired into a coherent overall model. In this talk, I will describe the major
components of the AEST and present some recent results constructed by the AEST.
Further information about the project and example results can be found from the RESOLV web
page, http://www.hhdc.bicc.com/resolv/
Automatic Extraction of Buildings from Aerial Images
C. Baillard, A. Fitzgibbon, C. Schmid, A. Zisserman
Dept.Of Engineering, University of Oxford
There is a recurring need for accurate, comprehensive 3D models of urban areas.
Applications include, map generation and update, placing of antennas for mobile
communications systems and town planning.
The traditional and well established solution to generating these 3D models is by the use
of photogrammetric techniques: typically a human operator works with two aerial images of
the urban area. The operator
identifies common features in the two images, such as a roof edge, and the 3D position of
that feature is then computed by triangulation.
The novelty in the work described here is that instead of two overlapping views, three
or more are used. These additional views have enabled successful automation of matching,
without requiring contextual information, and with almost no mismatches. The additional
views greatly strengthened geometric and photometric evidence for a match,
and also considerably improve the 3D accuracy because re-projection errors can be
minimized over all views.
Results are given for a state-of-the-art database of aerial images of residential areas in
Brussels.
Mobile Telepresence At Building Sites.
Gordon Mair
Strathclyde University
This talk will describe an EPSRC funded project to investigate the design and application
of hybrid virtual reality and telepresence systems for the planning and monitoring of
major engineering projects. Industrial partners
are Alvis, Babtie, and Orange. The construction industry is being used as an exemplar. The
system being developed comprises static and mobile stereo camera platforms. Novel aspects
of this project involve the use of
standard mobile phones to provide the control and video links to and from the vehicle and
cameras, the use of stereo co-encoding over the mobile phone system, and the real time
integration of real world and virtual world
images.
The role of vision in construction repair
Denis Chamberlain
Dept. Civil Engineering, City University.
Regarding the construction industry, there are, to date, few success stories in the
application of computer vision. There are a number of reasons for this, on both the
industry and technology-proposer sides.
On the industry side, there is currently little interest in hardware based research. What
little R&D resources exist, these are currently focused on increased profitability
through better IT. Entrepreneurial SME companies are the most likely developer-producers
of new equipment for the industry. However, the technology must perform
well in all respects, particularly important issues being profitability through use and
robustness against the harsh environment. There is also evidence of the burden of
miss-association e.g. a system that is part of
a non-existent or dubious host system.
A potentially useful system is a road crack repair system, currently being developed at
UTexas. Whilst it incorporates automatic crack detection, this is moderated on a
tele-operated basis, a measure that substantially
aids reliability. Another well founded development is a crane management system, where
image processing is used to identify packages. This is attractive because it aids site
safety and also enables billing disputes to be rapidly settled.