UTONOMOUS MODEL-BASED OBJECT IDENTIFICATION &

CAMERA POSITION ESTIMATION WITH APPLICATION TO

AIRPORT LIGHTING QUALITY CONTROL

James H. Niblock, Jian-Xun Peng, Karen R. McMenemy and George W. Irwin

Queen’s University Belfast, School of Electronics, Electrical Engineering and Computer Science

Ashby Building, Stranmillis Road, Belfast, N. Ireland, BT9 5AH

Keywords:

Model-based object tracking, camera position estimation.

Abstract:

The development of an autonomous system for the accurate measurement of the quality of aerodrome ground

lighting (AGL) in accordance with current standards and recommendations is presented. The system is com-

posed of an imager which is placed inside the cockpit of an aircraft to record images of the AGL during a nor-

mal descent to an aerodrome. Before the performance of the AGL is assessed, it is ﬁrst necessary to uniquely

identify each luminaire within the image and track it through the complete image sequence. A model-based

(MB) methodology is used to ascertain the optimum match between a template of the AGL and the actual

image data. Projective geometry, in addition to the image and real world location of the extracted luminaires,

is then used to calculate the position of the camera at the instant the image was acquired. Algorithms are

also presented which model the distortion apparent within the sensors optical system and average the camera’s

intrinsic parameters over multiple frames, so as to minimise the effects of noise on the acquired image data

and hence make the camera’s estimated position and orientation more accurate. The positional information is

validated using actual approach image data.

1 INTRODUCTION

Airport landing lighting has evolved from the early

beginnings when an aircraft’s relatively haphazard ar-

rival back to ground was conducted by the light of

gooseneck ﬂares. Rapid development took place be-

tween 1939 and 1945, when the Drem

system was

evolved; and subsequent post war developments in

omnidirectional low intensity approach lighting cul-

minated in the present state of the art, where high

intensity Calvert systems are the order of the day-or

rather night (Milward, 1976).

These lighting systems have evolved in the last

number of years in order to guide the pilot onto the

runway safely. Since the earliest days of ﬂying, pi-

lots have used ground references for navigation when

approaching an airport. Pilots need these visual aids

in good weather, as well as bad, and during the day

as well as at night (Horonjeff and McKelvey, 1993).

When visibility is poor and in night time conditions,

Drem

Lighting System was developed to assist Spitﬁre

landing in WW2

the visual information is reduced signiﬁcantly when

compared to the clear-weather daytime scene. It is

therefore essential to provide visual aids which are

as meaningful to pilots as possible (Horonjeff and

McKelvey, 1993). Today’s state of the art lighting

is referred to as aerodrome ground lighting (AGL)

and consists of approach lighting, elevated above the

ground, to guide the pilot onto the runway lighting

pattern and taxi the aircraft into its terminal. In or-

der to ensure the consistency of an airport lighting in-

stallation, strict guidelines are enforced on the posi-

tioning, uniformity, colour and intensity of the lumi-

naires

that make up the complete AGL.

The International Civil Aviation Organisation

(ICAO) has published a recommendation that the

measurement of luminous intensity, beam spread and

orientation of the luminaires, included in the approach

and runway lighting systems for a precision approach

complete lighting unit consisting of a lamp or lamps

together with the parts designed to distribute the light, to

position and protect the lamps and to connect them to the

power supply.

383

H. Niblock J., Peng J., R. McMenemy K. and W. Irwin G. (2008).

AUTONOMOUS MODEL-BASED OBJECT IDENTIFICATION & CAMERA POSITION ESTIMATION WITH APPLICATION TO AIRPORT LIGHTING

QUALITY CONTROL.

In Proceedings of the Third International Conference on Computer Vision Theory and Applications, pages 383-390

DOI: 10.5220/0001085503830390

 SciTePress

runway category I/II/III, should be undertaken using a

mobile measuring unit of sufﬁcient accuracy to anal-

yse the characteristics of the individual luminaires.

To assess the performance and alignment of the lu-

minaires it is necessary to know more about the AGL

and how the luminaires are arranged. AGL consists of

runway lighting separated from the approach lighting

system (ALS) by a row of luminaires termed the run-

way threshold. Runway and threshold luminaires are

usually inset, that is, they are installed at ground level

whilst the approach luminaires are elevated above the

ground. A typical AGL (CATI (ICAO, 2004)) layout

is illustrated in ﬁgure 1.

Figure 1: AGL Layout.

This paper presents results from research con-

ducted into creating an aerial-based vision system ca-

pable of autonomous performance assessment of the

complete AGL pattern. The work proposes mounting

one, or more, cameras in the aircraft, capable of ac-

quiring image data of a descent to the airport as the

aircraft performs a landing. The function of the cam-

era is to replicate what the pilots see during a standard

approach and store the information to an external hard

drive for off-line performance assessment.

To assess the performance of luminaires a number

of processes need to be undertaken. Firstly, the noise

apparent in the camera needs to be quantiﬁed in the

form of a distortion matrix. The next problem is that

of uniquely identifying each luminaire from the ac-

quired image data. Niblock et al. compared a basic

single pixel image-based tracking method against ex-

isting tracking techniques, such as, the KLT and SIFT

alternatives (Niblock et al., 2007a). This work high-

lighted the limitations of such a tracking system and

proposed that in order to uniquely identify each lumi-

naire, and thus assess its performance, a model-based

approach is required.

In this paper the model-based tracker is brieﬂy dis-

cussed before showing how this technique was up-

dated for the purposes of camera position and pose

determination. The model-based approach attempts

to match a template of the ALS to the set of extracted

luminaires from the image, where the template of the

ALS is illustrated in ﬁgure 2. Strict standards en-

forced by the ICAO (ICAO, 2004) are in place for the

positioning, uniformity, projection angle and colour

of these luminaires.

The major advantage of this approach, is that for

each successfully matched luminaire, information re-

garding its position is known, both within the image

and real-world coordinate frames. Therefore, projec-

tive geometry can be utilised to estimate the camera

position and orientation data at the instant each image

was acquired. It is essential in this work to have ac-

curate camera position and pose information for the

luminaire performance assessment. Existing research

indicates that an image of a luminaire can be used to

estimate the intensity of a luminaire, providing accu-

rate position and orientation of the camera in relation

to the luminaire is known (McMenemy, 2003). The

contribution of the paper is to present a model-based

methodology for luminaire identiﬁcation which is ex-

tended in order to estimate the position and orienta-

tion of the camera (aircraft) during its descent. The

theory is validated using actual approach data before

concluding remarks and future work are discussed.

2 MODEL-BASED (MB)

TRACKING

In order to assess the performance of the luminaires

it is ﬁrst necessary to uniquely identify each lumi-

naire in the ALS and track them through an image

sequence in order to build up a proﬁle of the lighting

pattern. Once the luminaires have been identiﬁed and

labelled, these features are then used to estimate the

camera’s position during the landing. It is essential

that the estimated camera position is accurate and ro-

bust to noise as any inaccuracies will have a follow on

effect with the performance assessment software.

The objective of the MB tracker is one of trying

to match a template which consists of the luminaires

within the ALS to a set of features extracted from im-

age data acquired during the approach to the airport.

VISAPP 2008 - International Conference on Computer Vision Theory and Applications

384

Figure 2: Standard Category I (CATI) approach lighting

pattern.

This process is known as planar homography and con-

sists of ﬁnding a one-to-one match between a pair of

images (Vincent and Laganiere, 2001) or, in this case,

a template of the ALS and an image.

In order to project this template onto the acquired

image data a pinhole camera projection system is

utilised which is mathematically modelled as consist-

ing of a projection system and an imaging system.

The projection system projects a 3D-object from the

real world (i.e. an ALS luminaire) to the image plane

of the camera. The imaging system converts the pro-

jected 2D-object on the image plane into a series of

pixels through an image sensor array.

According to the pinhole imaging principle de-

picted in ﬁgure 3 a 3D-point (x, y, z) expressed in the

camera coordinate system OXY Z is projected to a 2D-

point (u,v) on the image plane using a projection sys-

tem. From ﬁgure 3 we can summarise (1).

u = f x

−1

v = f x

−1

(1)

where f denotes the focal length of the camera. It

is necessary to allow the focal length of the cam-

era to vary so that the camera will be capable of au-

tonomously locating the AGL and zooming in on the

relevant area of the lighting pattern, for example the

ALS. This process enhances the quality of the ac-

quired image data.

Assuming that the image sensor array is evenly

conﬁgured as orthogonal grids, that is to say a se-

Figure 3: Model-based projection system.

ries of rectangular-shaped pixels aligned in a grid-

like structure in the image plane, the imaging system

can then be represented as the linear transformation

shown in equation (2),

p = k

+ k

u = k

+ α

−1

q = k

+ k

v = k

+ α

−1

(2)

where (

q) denotes the coordinates of the pixel cor-

responding to the 2D-point projection (u,v) in the im-

age plane given in (1); k

and k

are the number of pix-

els per unit distance in the u and v directions respec-

tively, and α

= k

f , α

= k

f represent the intrin-

sic camera parameters, where f represents the focal

length of the imaging system and k

represent the

scaling factor of the image data and the horizontal and

vertical image plane respectively. Finally, (k

, k

)

represent the image coordinates of the intersection of

the optical axis and the image plane, henceforth re-

ferred to as the principal point.

However, the approaching template is normally

deﬁned with the 3D coordinates of the 120 luminaries

given in the airport coordinate system O

. The

transformation from the airport coordinate system to

the camera coordinate system includes a translation

and a rotation as shown in equation (3).









= R(γ, θ, ψ)





− x

− y

− z





(3)

where the translation o = [x

, y

, z

]

contains the co-

ordinates of the camera position (deﬁned at the orig-

inal point of the camera coordinate system, see ﬁg-

AUTONOMOUS MODEL-BASED OBJECT IDENTIFICATION & CAMERA POSITION ESTIMATION WITH

APPLICATION TO AIRPORT LIGHTING QUALITY CONTROL

385

ure 3) in the airport coordinate system. The vec-

tor t = [x

, y

, z

]

contains the coordinates of an ap-

proaching luminaire in the airport coordinate system

and (γ, θ, ψ) denote the yaw, pitch and roll of the cam-

era system respectively (i.e. the three rotations around

the X,Y and Z axis respectively). Where R(γ, θ, ψ) is

the corresponding rotation matrix, which is formed

by three rotations around the Z, Y and X axes respec-

tively of the camera coordinate system, as shown in

equation (4),

R(γ, θ, ψ) = R

(γ)R

(θ)R

(ψ)

(γ) =





1 0 0

0 cosγ sinγ

0 −sin γ cosγ





(θ) =





cosθ 0 −sinθ

0 1 0

sinθ 0 cosθ





(ψ) =





cosψ sinψ 0

−sin ψ cosψ 0

0 0 1















(4)

where R

(ψ), R

(θ) and R

(γ) are the corresponding

rotation matrices around Z, Y and X axes, respectively

(V. Lepetit, 2005).

Using the pin-hole camera projection system il-

lustrated in ﬁgure 3 the template of the ALS is su-

perimposed onto the image data and the Levenberg-

Marquardt (LM) method used to minimise the error

between the two sets of data. This procedure is de-

tailed by Niblock et al. (Niblock et al., 2007b) and

Peng et al. (Peng et al., 2006). Niblock et al. show

that this process works well and results in successfully

identiﬁed luminaires in the image sequence (Niblock

et al., 2007b).

The major advantage offered by a model-based

methodology is its ability to identify luminaires that

are missing (or have been turned off) in the ALS.

Techniques such as KLT and SIFT only track lumi-

naires that are present in the image data. It is essential,

for the performance assessment work, that if a lumi-

naire is missing, for any reason, that its position is still

recorded and its associated grey level stored. By us-

ing a template of the ALS this is made possible. Fur-

thermore, the model-based methodology produces the

best results on the actual image data (Niblock et al.,

2007b). As the noise level is increased, it is essen-

tial that the algorithms are more robust and have a

high tolerance level to noise inherent within the image

data. If this is not the case false matches can be made

and the grey level proﬁles of the extracted luminaires

can become confused, which is highly undesirable for

this application.

These identiﬁed luminaires can then be used to es-

timate the position and orientation of the camera at

the instant the image was taken, which as previously

mentioned is essential for luminaire performance as-

sessment.

2.1 Camera Positioning

Being able to estimate position and orientation infor-

mation from image data is a well researched area in

computer vision. Indeed, work has already been con-

ducted in the area of aircraft positioning by Soni et al.

(Soni and Sridhar, 1994) and Sridhar et al. (Sridhar

et al., 1996; Chatterji et al., 1998) who produced sys-

tems that utilise the information provided by the posi-

tion of individual luminaires in an image for estimat-

ing the relative position and orientation of an aircraft.

The location of luminaires within the image plane is

derived using perspective projection equations based

on a pinhole camera model (Faugeras and Toscani,

1987). Differences of features tracked between suc-

cessive images are used in conjunction with a recur-

sive optimisation algorithm in order to ﬁnd the op-

timum position and orientation of the aircraft. The

drawback of Soni’s work is that the roll of the air-

craft is obtained using a roll sensor and is not ob-

tained, like the other variables, from the image se-

quence. Mostafavi et al. use similar techniques with

external information such as Differential Global Posi-

tioning System (DGPS) data in order to ascertain the

position and orientation of the camera with relation to

the runway outline and markers (Mostafavi and Mal-

one, 1995).

The work presented in this paper differs from the

aforementioned research by presenting novel tech-

niques which average the intrinsic camera parame-

ters over multiple frames, in order to minimise the

effects of noise inherent within the image data, and

thus make the estimated extrinsic parameters more ac-

curate. A further contribution of this work is that a

varying focal length may be used by the imaging sys-

tem is order to acquire image data of the highest qual-

ity, which is essential for the performance assessment

of the ALS. The previous work in this area, gener-

ally assume constant intrinsic parameters. To realise

an accurate imaging system, distortions caused by the

sensors optical system also need to be modelled and

accounted for.

2.2 Distortion Correction

The model outlined in section 2 makes a number of

assumptions regarding the optical system utilised by

the imaging system. If the acquired image data has

VISAPP 2008 - International Conference on Computer Vision Theory and Applications

386

been affected by distortion the image can change in

shape and thus the points are no longer assigned to the

pixel position that was estimated using the previous

model.

That is to say, for real cameras, a pure perspec-

tive projection model is not always sufﬁcient to pre-

cisely represent the image data. Several types of im-

perfections in the design, machining and assembly of

the camera optical (lens) system may result in small

geometrical distortion (Heikkila, 2000). Most com-

monly the distortion errors are decomposed into ra-

dial (ﬁgure 4(c)) and tangential (ﬁgure 4(b)) compo-

nents (Faugeras and Toscani, 1987), where the perfect

undistorted image is illustrated in ﬁgure 4(a). For the

distorted image coordinates given by (r

, c

), the cor-

rected (or undistorted) coordinates are expressed as

equation (5).

= c

+ δ

, c

)

= r

+ δ

, c

)

(5)

where (r

, c

) denote the distortion-free image co-

ordinates (that are unobservable) corresponding to

, c

) that are observable in the actual image, and

, c

), δ

, c

) denote the total displacements

along the column and row directions respectively.

Heikkila gives further details on image distortion and

the relevant tangential and radial distortion models

(Heikkila, 2000). Utilising these standard techniques

the distortion in the image data is corrected for, be-

fore any estimates of the camera position and orien-

tation are calculated. In addition to distortion correc-

tion, it is important to realise that the focal length of

the camera can vary during the approach. A change in

the focal length will obviously lead to a change in the

projection model, and thus it is necessary to account

for this. The algorithms to do this are now discussed.

2.3 Multi Frame-based Estimation

The objective of the multi frame-based estimation

technique is to create an algorithm that allows for a

varying focal length. The focal length is one of the

intrinsic camera parameters modelled in equation (2),

represented by the α

, α

coefﬁcients. To do this, the

technique assumes that the intrinsic camera parame-

ters remain constant for a predeﬁned time period, e.g.

1 second. Therefore, this work estimates the intrin-

sic camera parameters over a predeﬁned number of

images and averages them in order to minimise error

caused by noise inherent within the acquired image

data. These optimised intrinsic camera parameters are

then used to estimate the camera’s position and orien-

tation (i.e. extrinsic parameters) during the landing.

Suppose a sequence of F

images are as-

sessed and the intrinsic parameters and distor-

tion coefﬁcients of the camera are assumed con-

stant over the F

frames, denoted as w

int

, k

, α

, k

, t

]

, where w

int

represents

the intrinsic camera parameters, where (k

, k

) de-

notes the principal point of the image, (α

, α

)

are a combination of the focal length and res-

olution of the camera and the k, t coefﬁcients

represent the radial and tangential distortion pa-

rameters respectively, as modelled by Lepetit et

al. (V. Lepetit, 2005). If the camera ex-

trinsic parameters for frame f

are denoted as

ext

( f

) = [t

( f

), t

( f

), t

( f

), ψ( f

), θ( f

), γ( f

)]

where (t

( f

), t

( f

), t

( f

)) represent the camera’s po-

sition in a given frame and (ψ( f

), θ( f

), γ( f

)) the

orientation, then the cost function is deﬁned in equa-

tion (6) as,

E(w

ext

, w

int

) =

∑

k=1

( f

;k)+ e

( f

;k)] (6)

where k = 1, ..., N and represents the number of ex-

tracted luminaires in the image. Equation (6) thus

needs to be minimised with respect to w

int

and

ext

( f

) for f

= 1, ..., F

( f

, k) = c(k)(w

int

, w

ext

( f

)) − c

( f

, k)(w

int

)

( f

, k) = r(k)(w

int

, w

ext

( f

)) − r

( f

, k)(w

int

)

(7)

where r(k)(w

int

, w

ext

( f

)) and c(k)(w

int

, w

ext

( f

)) de-

note the projection coordinates of the k

ALS lumi-

naire produced in equation (8),

c = α

−1

+ k

r = α

−1

+ k

(8)

which are functions of the intrinsic parameters w

int

and the extrinsic parameters w

ext

( f

) for frame

; r

( f

, k)(w

int

) and c

( f

, k)(w

int

) denote the

distortion-free image coordinates of the blob corre-

sponding to the k

luminaire extracted from frame

, which are functions of the intrinsic parame-

ters w

int

, or more speciﬁcally, functions of part

, k

, t

) of w

int

With an optimised multiple frame-based camera

parameter estimation now established, section 2.4 de-

tails the constraints applied to the software. Once

these constraints are deﬁned the software is tested us-

ing actual airport lighting data.

2.4 Constraints

A number of constraints are placed upon the camera

positioning algorithms. When making a normal 3 de-

gree approach to the airport a number of assumptions

AUTONOMOUS MODEL-BASED OBJECT IDENTIFICATION & CAMERA POSITION ESTIMATION WITH

APPLICATION TO AIRPORT LIGHTING QUALITY CONTROL

387

(a) (b) (c)

Figure 4: (a) Original image (b) Tangential distortion (c) Radial distortion.

Figure 5: Camera position and orientation for actual ap-

proach data using the multi frame-based estimation.

can be made. Firstly the camera starts to acquire im-

age data from roughly 2.5km from the airport. If a 3

degree approach is assumed this means that the height

of the aircraft with respect to the AGL is approxi-

mately 131m. Therefore, boundaries can be placed

on the X,Y and Z data, for example 2500±1000m,

0±50m and 150±150m respectively. However, if it

is not possible to assume a 3 degree approach and a

different angle of approach is used, then trigonometry

can be used to update the starting constraints and their

respective upper and lower limits. A second assump-

tion is that the ﬁeld of view of the camera is set to 45

degrees. This assumption helps with the segmentation

of the ALS and ensures the noise from surrounding

light sources and general background illumination is

kept to a minimum.

With the optimised intrinsic camera parameters

estimated over multiple frames, the next section val-

Figure 6: Error proﬁle for the camera’s position and orien-

tation for actual approach data using the multi frame-based

estimation.

idates the new software using actual airport lighting

data acquired during an approach to Belfast Interna-

tional Airport.

3 POSITION & ORIENTATION

RESULTS

During a complete ﬂight test a number of approaches

are made to the airport. The vision system detailed in

this paper was mounted in the cockpit of the aircraft

and set to acquire image data during these approaches.

The algorithms detailed were used to uniquely iden-

tify each of the luminaires in the ALS before using

this information to estimate the aircraft’s position and

orientation during the approach. To assess the accu-

VISAPP 2008 - International Conference on Computer Vision Theory and Applications

388

Figure 7: Missing/merged luminaires (left image) and optimisation results (right image). Note how the luminaires are still

identiﬁed by the model-based matching algorithm even though they are not present in the image data. (Niblock et al., 2007b).

racy of the proposed algorithms, they are compared

to the Flight Precision’s ILS position and orientation

data.

The successfully extracted luminaires were then

used to estimate the camera’s position and pose. The

model-based results for a sample image are shown

in ﬁgure 7. The dots represent extracted luminaires

present in the image data, using the connected com-

ponent analysis technique outlined in (Niblock et al.,

2007a). Most luminaires have been successfully iden-

tiﬁed and these are represented by the dot with a cir-

cle around it. However, a number of the circles have

a cross (‘+’) inside them, which indicates a luminaire

that is missing or hasn’t been extracted from the im-

age data. These can be caused by merged luminaires

(illustrated at the top of the image), or because the

luminaires are actually missing or occluded (as illus-

trated in the bottom right of the ﬁgure where 5 consec-

utive luminaires are occluded by a temperature meter

housed outside the aircraft).

Figure 5 shows how the camera parameters es-

timated using the multi frame-based software, com-

pared with the Flight Precision ILS data. The re-

sults show a good comparison in terms of the esti-

mated camera position. There is also a good corre-

spondence between the pose information. More im-

portantly there is a strong correlation between the pro-

ﬁles, with a constant offset apparent between the two

sets of position and pose data illustrated in ﬁgure 5.

The reason for this offset (and why both sets of data

are not superimposed on top of each other) is that the

ILS data shows the aircraft’s position and pose with

respect to the PAPI luminaires (illustrated in ﬁgure 2),

whereas the model-based positioning software calcu-

lates the position and pose of the camera with respect

to the centre of the threshold.

The error proﬁle, which is the difference between

the two sets of data, is shown in ﬁgure 6. The ﬁg-

ure shows that the positional error is largest for the

X parameter. Note that the errors shown in ﬁgure 6

include the constant difference between the two ref-

erence systems of the ILS and image-based data. For

example, with the camera pose information, the er-

ror between the two sets of data is negligible with the

highest error set at 5 degrees which is explained by the

two difference reference coordinate frames utilised. It

is also worth highlighting that the error proﬁle is min-

imised because the camera’s intrinsic parameters are

averaged over multiple frames thus causing the error

of the camera’s extrinsic parameters (caused by fac-

tors such as stray noise) apparent in any given image

to be minimised.

4 CONCLUDING REMARKS

This paper presents results from research conducted

into creating an aerial-based vision system capable of

autonomous performance assessment of the complete

AGL pattern. The work proposes mounting one, or

more, cameras in the aircraft, capable of acquiring im-

age data during a typical descent to the aerodrome as

the aircraft performs a landing.

To date algorithms have been produced in order

to robustly extract the luminaires present within the

image data and identity them using a model-based

methodology. A pin-hole camera projection system

was then used to estimate the position and orienta-

tion of the camera, during the descent, from the ac-

quired image data. The results obtained from the new

software were compared against the positional infor-

mation supplied by the Flight Precision ILS data and

a strong correlation was found between both sets of

data. In particular the proﬁles of the two sets of posi-

AUTONOMOUS MODEL-BASED OBJECT IDENTIFICATION & CAMERA POSITION ESTIMATION WITH

APPLICATION TO AIRPORT LIGHTING QUALITY CONTROL

389

tional and orientation data were found to have a strong

correlation with the constant error offset explained by

the different reference systems utilised by both tech-

niques.

The ﬁnal goal of this work is to realise an au-

tonomous image processing system capable of as-

sessing the performance of the complete AGL pat-

tern. Thus far, MATLAB has been used for the soft-

ware with a mean execution time per frame of ap-

proximately 3 seconds and a standard deviation of

0.3s/frame achieved using a standard CPU with a

3GHz processor and 1GB RAM. This timing infor-

mation could be dramatically reduced with a C++

setup and further reduced if the algorithms were pro-

grammed with a GPU as implemented by Sinha et al.

(Sinha et al., 2006).

Future work includes assessing the performance

of the lighting pattern. To this end two methodolo-

gies are proposed. Firstly, uniformity, which assesses

the performance of the complete lighting pattern and

secondly, assessing the luminous intensity of each in-

dividual luminaire within the lighting pattern. This

work will add negligible time onto the execution time

per frame as the memory inefﬁcient and time con-

suming tasks of reading in an image sequence and

extracting the luminaires/eatimating the aircraft’s po-

sition have already been determined. Thus, this paper

shows that it is possible to assess the performance of

the AGL using an aerial-based imaging methodology.

ACKNOWLEDGEMENTS

The authors would like to thank the EPSRC (Grant:

EP/D05902X/1) for their ﬁnancial backing for the

research (2004-2007) and the Royal Academy of

Engineering for their postgraduate fellowship award

(2005). Furthermore, we would also like to acknowl-

edge the contribution of Flight Precision and Belfast

International Airport for providing ﬂight time in order

to collect airport lighting data and test the developed

algorithms.

REFERENCES

Chatterji, G., Menon, P., and Sridhar, B. (1998). Vision-

based position and attitude determination for aircraft

night landing. Journal of Guidance, Control, and Dy-

namics, 21(1).

Faugeras, O. and Toscani, G. (1987). Camera calibration

for 3d computer vision. Proc. Int’l Workshop Indus-

trial Applications of Machine Vision and Machine In-

telligence, pages 240–247.

Heikkila, J. (2000). Geometric camera calibration using cir-

cular control points. IEEE Trans. PAMI, 22(10).

Horonjeff, R. and McKelvey, F. (1993). Planning and De-

sign of Airports. McGraw Hill, 4th ed. edition.

ICAO (2004). Aerodrome Design and Operations. Interna-

tional Civil Aviation Organization, 4th edition.

McMenemy, K. (2003). Photometric Evaluation of Aero-

drome Ground Lighting. PhD thesis, Queen’s Univer-

sity Belfast.

Milward, R. (1976). New approach to airport lighting in-

spection. Shell Aviation News, 437:26–31.

Mostafavi, H. and Malone, M. (1995). Landing trajectory

measurement using onboard video sensor and runway

landmarks. Proceedings of SPIE - The International

Society for Optical Engineering, 2463:116–127.

Niblock, J., Peng, J., McMenemy, K., and Irwin, G.

(2007a). Autonomous tracking system for airport

lighting quality control. Proceedings of the 2nd Inter-

national Conference on Computer Vision Theory and

Applications, VISAPP, Motion Tracking and Stereo

Vision:317–324.

Niblock, J., Peng, J., McMenemy, K., and Irwin, G.

(2007b). Fast model-based feature matching tech-

nique applied to airport lighting. Transactions of the

IET Science, Measurement & Technology, In Press.

Peng, J., Li, K., and Huang, D. (2006). A hybrid forward

algorithm for rbf neural network construction. IEEE

Transactions on Neural Networks, 17(6):1439–1451.

Sinha, S., Frahm, J., Pollefeys, M., and Genc, Y. (2006).

Gpu-based video feature tracking and matching. Tech-

nical report 06-012, UNC Chapel Hill Department of

Computer Science.

Soni, T. and Sridhar, B. (1994). Modelling issues in vi-

sion based aircraft navigation during landing. IEEE

Workshop on Applications of Computer Vision Pro-

ceedings, pages 89–96.

Sridhar, B., Chatterji, G., and T.Soni (1996). Model-based

vision for aircraft position determination. Control En-

gineering Practice, 4(8):1153–1159.

V. Lepetit, P. F. (2005). Monocular model-based 3d tracking

of rigid objects: A survey. Foundations and Trends in

Computer Graphics and Vision, 1(1):1–89.

Vincent, E. and Laganiere, R. (2001). Detecting planar ho-

mographies in an image pair. IEEE Proceedings of

the 2nd International Symposium on Image and Sig-

nal Processing and Analysis, pages 182–187.

VISAPP 2008 - International Conference on Computer Vision Theory and Applications

390