Sandau R. Digital Airborne Camera: Introduction and Technology

Подождите немного. Документ загружается.

1.2 Applications for Digital Airborne Cameras in Photogrammetry and Remote Sensing 11

Fig. 1.2-3 Illustration of a ﬁlter design mainly for observing vegetation

vegetation and is also used to detect chlorophyll in water bodies. The second absorp-

tion band of chlorophyll lies in the 635 ± 25 nm red s pectral channel (maximum at

650 nm).

At 860 ± 25 nm, the NIR channel lies on the plateau of vegetation curves and, in

conjunction with the red channel, which ends in front of the vegetation edge (“red

edge”), provides data on the state of the vegetation.

The spectral responsivity of colour ﬁlm is broad-banded for the various spectral

channels, and the spectral channels overlap (Fig. 1.2-4). This overlap of the spectral

Fig. 1.2-4 Spectral responsivity of colour ﬁlm material and of a detector system designed for

photogrammetry and remote sensing (Leica, 2004)

12 1 Introduction

channels is good for the colour ﬁlm impression or infrared colour ﬁlm impression,

but it does not support remote sensing applications. The central wavelengths and the

spectral bandwidths for multispectral and true-colour images differ. Multispectral

applications require non-overlapping, narrow spectral bands, and, for observing veg-

etation, also an infrared channel near the vegetation edge (red edge). In contrast, the

true-colour channels are rather more adjusted to spectral visual sensitivity and for

this reason are broad-banded and overlapping.

Hence, a sensor system or digital airborne camera for photogrammetry and

remote sensing must be equipped with ﬁlters that generate narrow spectral channels

which are separated from each other. The true-colour images can be derived from

the RGB channels designed in this manner using a process of colour transformation,

possibly in conjunction with a panchromatic channel (see Section 2.7).

The absorption ﬁlters mounted during the manufacture of CCD matrices are also

not so well suited to the multispectral applications discussed above.

The bandpass response of typical narrow-band absorption ﬁlters is shown in the

right-hand part of Fig. 1.2-5. Absorption ﬁlter Absorption ﬁlters cannot be produced

as perfectly and with such narrow bands as interference ﬁlters (see Section 4.3.2).

The transitions are not sufﬁciently steep. For instance, the green ﬁlter does not com-

pletely absorb the red and blue parts of the spectrum. The interference ﬁlters shown

on the left-hand sides of Figs. 1.2-4 and 1.2-5 can be implemented with much greater

precision. It should be mentioned, however, that a considerable amount of effort is

involved in manufacturing them (many metal oxide layers have to be vapour-plated

on to the glass bases in a vacuum).

Summing up, it can be said that modern digital detector components enable dig-

ital airborne cameras for photogrammetry and remote sensing to be developed and

manufactured. Often they make it possible to improve the quality of the results

achieved using analogue airborne cameras and to expand the range of applications.

Fig. 1.2-5 Comparison of spectral responsivities of detectors ﬁtted with absorption ﬁlters and a

detector system designed for photogrammetry and remote sensing (Leica, 2004)

1.3 Aircraft Camera or Satellite Camera 13

1.3 Aircraft Camera or Satellite Camera

Digital airborne sensors suitable for photogrammetry and remote sensing applica-

tions currently occupy a position between analogue airborne cameras, which poten-

tially have higher geometric resolution but limited spectral variability, dependent

on the available ﬁlm material, on the one hand, and satellite systems, which have a

lower geometric resolution, but in many instances a higher spectral resolution, on

the other.

Figure 1.3-1 is a schematic representation of the performance of sensor sys-

tems based on the key parameters: geometric and spectral resolution. Analogue

airborne cameras can provide virtually any resolution down to about 1 cm. Digital

airborne cameras are already capable of a ground sample distance (GSD) below

5 cm. Satellite systems using panchromatic systems (roughly comparable to black-

and-white ﬁlm) have reached a GSD below 0.5 m. Figure 1.3-2 illustrates the

development of GSD of satellite systems. From 80 m GSD achieved by ERTS (later

renamed Landsat-1) – the ﬁrst satellite launched for civil earth observation in 1972 –

we have advanced to 0.41 m GSD.

Fig. 1.3-1 Performance for sensor systems

Table 1.3-1 shows characteristics of selected satellite systems, such as the coarser

GSD of multispectral channels, swath width, scene size and revisit time (time inter-

val until the next opportunity to photograph the same area), indicating the potential

geometric and time coverage.

1.3 Aircraft Camera or Satellite Camera 15

Fig. 1.3-3 Connection between repeat rate and ground resolution (Konecny, 2003)

can supply images of intermediate resolution (5–14 m pan and 10–28 m MS) only a

few times a year. But coverage of large areas is possible only at intervals of several

years. High-resolution satellite systems (IKONOS with 1 m GSD or Russian pho-

tographic systems with a resolution of 2 m) are approaching the resolution range

of aerial images obtained from high-ﬂying aircraft. The highest resolutions in the

decimeter to centimeter range are obtained with low-ﬂying airborne cameras or

ground surveys at correspondingly longer time intervals.

Figure 1.3-4 illustrates the huge differences in altitude from which civil earth

observation systems and aircraft systems carry out imaging operations. The differ-

ence of about 1:200 (3–600 km) illustrates, in qualitative terms at least, that satellite

systems are much more complex and costly to make than aircraft systems to achieve

comparable GSDs. This is also reﬂected in the costs of the image products and

explains why more than 80% of the earth is mapped with airborne cameras. The

main reason lies in the ratio between the focal lengths, which, like the altitude ratio,

is 1:200. Figure 1.3-4 gives an impression on the ﬂight hight ranges of airborne and

spaceborne sensors. Speed differences also play a role. An aircraft, for instance, ﬂies

at a speed of 70 m/s, whereas the ground track of an LEO satellite at an altitude of

roughly 600 km is approximately 7 km/s. The speed ratio and hence the ratio of the

integration times is 1:100.

16 1 Introduction

Fig. 1.3-4 General situation

for ﬂighing hight of airborne

and space borne sensors

1.3.1 Detection, Recognition, Identiﬁcation

The information the human brain is able to derive from an image depends on the

number of picture dots (pixels) in a pixel conglomerate. Image interpreters typically

use the following decision levels:

• Detection: discovering the existence of an object

• Recognition: classifying the object as belonging to a type group

• Identiﬁcation: identifying the object type.

The number of pixels occupied by the object determines the certainty of the deci-

sion regarding the existence of an object or even its identiﬁcation. Information about

the object type is not substantially improved, nor the likelihood of identiﬁcation of

a certain object type substantially increased, if, for instance, a the number of pixels

viewed is a 100 times the number required to reach a decision. In other words, the

object’s size and structure determine the required pixel size or GSD. In this respect,

there are optimal applications for high-resolution sensors with small GSDs and for

sensors with larger GSDs. The critical question is how many pixels are needed to

achieve a certain decision quality. Figure 1.3-5 illustrates the issue sketched in the

foregoing.

Figure 1.3-6 is an example in which the identiﬁcation of a car about 5 m in length

is used to illustrate the relationships. The component pictures in Fig. 1.3-7 which

are to be recognised and identiﬁed are once again highlighted.

Figure 1.3-8 shows the situations in which decisions would have to be made when

viewing the image material obtained by the QuickBird satellite with GSD = 0.8 m,

1.3 Aircraft Camera or Satellite Camera 17

Fig. 1.3-5 GSD and object identiﬁcation (Leica, 2004)

Fig. 1.3-6 GSD and object identiﬁcation using a car about 5 m long (Leica, 2004)

a digital airborne camera with GSD = 0.2 m and an analogue airborne camera with

a simulated GSD = 0.1 m.

The number of object pixels required for the three decision levels is not consis-

tently deﬁned. Practitioners sometimes refer to the Johnson criteria (Johnson, 1985),

which have been modiﬁed somewhat over the years. Table 1.3-2 shows the number

of object pixels required for a decision probability of 50% for the object’s one- and

two-dimensional expansion (Holst, 1996).

18 1 Introduction

Fig. 1.3-7 Component images for reconition and identiﬁcation from Fig. 1.3-5 as an illustration

(Leica, 2004)

Fig. 1.3-8 Images from

QuickBird, GSD GSD = 0.8

m, a digital airborne camera,

GSD = 0.2 m, and an

analogue airborne camera,

simulated GSD = 0.1 m

(Leica, 2004)

1.3 Aircraft Camera or Satellite Camera 19

Table 1.3-2 Decision thresholds derived from the Johnson criterion

Decision Description

One-dimensional

2D N

Detection The pixel represents an object being

sought with good probability

10.75

Recognition Allocation to a speciﬁc class with

hierarchical certainty

Identiﬁcation Speciﬁcation within a class with

hierarchical certainty

Table 1.3-3 Factors for

changing the outcome

probabilities in Table

1.3-2

Outcome probability Factor

1.00 3.0

0.95 2.0

0.80 1.5

0.50 1.0

0.30 0.75

0.10 0.50

0.02 0.25

Table 1.3-3 shows the requisite factors by which the values in Table 1.3-2 need

to be multiplied to obtain an outcome probability other than 50% (Ratches, 1975).

It is contended, therefore, that aircraft and satellite systems are complementary.

Some essential characteristics of satellite systems are:

• Fixed orbit: area coverage being predictable but dependent on cloud cover

• Applicable without having to invest a great deal of effort into preparation

• Fixed GSD, current minimum 0.5 m panchromatic and 2 m multispectral

• Known costs per scene.

Essential characteristics of aircraft systems:

• Flexible application on demand

• Usable even in unfavourable weather (ﬂying below the clouds)

• GSD adaptable to the task at hand by varying ﬂight altitude

• Stereo data easily acquired.

Data from airborne and satellite sensors is complementary in many applications.

Different sensors and platforms have to be used owing to the great variety of objects

to be observed and/or surveyed with respect to size, shape, texture and colour or to

properties that can be differentiated only with multispectral or hyperspectral sensors.

Given the increasingly expanding ﬁelds of application and the use of GIS stations,

it is necessary to fuse all types of sensor date in order to meet the demand for

information quickly and reliably. Data fusion and GIS have a high potential for

creating new lines of business.

20 1 Introduction

1.4 Matrix Concept or Line Concept

The technological possibilities, potential quality expectations and possible applica-

tions in photogrammetry and remote sensing discussed in Sections 1.1, 1.2, and

1.3 apply to digital airborne cameras regardless of any speciﬁc design concept.

The extent to which possibilities become concrete solutions, however, depends on

the basic underlying concept, based on either detector lines or detector matrices.

Extensive variations are, of course, possible within the basic concepts, some of

which are discussed in Section 1.5.

The key difference between the basic concepts is shown in Fig. 1.4-1. The line

variant generates long, continuous strips of images, whereas the matrix variant gen-

erates rectangular, in most cases square, images that can be joined to form image

strips. Matrix-based cameras are used like analogue ﬁlm cameras to implement

stereoscopic applications with overlaps such as 60% in the direction of ﬂight. In

most cases, the three-line concept is used in line-based airborne cameras (Hofmann,

1988). Figure 1.4-2 shows a schematic representation of the procedure.

In the case of the three-line camera, all objects on the ground are photographed

from three different directions. This redundancy results in robust solutions in tri-

angulation or in creating DEMs (digital elevation models). In photographs taken

with a matrix camera at 60% forward overlap, 60% of the objects appear in three

photographs.

In the line camera, the convergence angles are a function of the distance d from

the nadir line

Fig. 1.4-1 Sample systems exemplifying the line concept and the matrix concept (Leica, 2004)