The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part B4. Beijing 2008
406
For applications, where the intensity is of interest, a
normalization based on the incidence angle is needed. The
variation of the incidence angle increases if data from several
flights with different paths are fused.To give an example a
RGB-image together with the corresponding intensity values
from two different flights is visualized in
Figure 1. The viewing direction of the sensor system is depicted
by a black arrow. The area of interest is the gabled roof. The
roof area orientated towards the sensor systems delivers higher
intensity values, while the turn away roof area delivers
significant smaller intensity values.
Figure 1: Dependency of the intensity from the incidence angle:
a) RGB image,
b) Intensity values of flight 3,
c) Intensity values of flight 4.
These from the waveform estimated values are strongly
correlated to the incidence angle of the laser beam on the
surface. Therefore we propose to normalize the value of the
intensity by considering the incidence angle derived by the
sensor and object position as well as its surface orientation.
In Section 2 a brief description of the used full-waveform data
is given. Further we introduce the physical constraints, the
required normalization step, and the methodology for
calculation of the normal vectors of the surfaces based on the
covariance matrix and the resulting incidence angle. The
description of the scene and the gathered data is presented in
Section 3. In Section 4 homogenous test regions are selected for
the assessment of the normalization. The results including a
presentation of the values before and after the normalization are
demonstrated in Section 5. Finally the used formula and derived
results are discussed.
2. METHOLOGY
2.1 Terminology of Full-Waveform Data
By the full-waveform laser data acquisition for each beam the
total number of detected backscattered pulses is known and is
assigned to the corresponding echoes. Each echo is described
by a point with its 3d coordinate, signal amplitude a , and
signal width w at full-width-at-half-maximum derived from the
Gaussian approximation. Additionally the 3d coordinate of the
sensor position is available.
The shape of the received waveform depends on the illuminated
surface area, especially on the material, reflectance of the
surface and the inclination angle between the surface normal
and the laser beam direction. The typical surface attributes
which can be extracted from a waveform are range, elevation
variation, and reflectance corresponding to the waveform
features: time, width and amplitude.
The intensity (energy) is estimated by the width multiplied with
the amplitude of the Gaussian approximation and modified by
the range between sensor and object with respect to the
extinction by the air. It describes the reflectance influenced by
geometry and material of the object at this point. For each
particular echo caused by partially illuminated object surfaces
we receive an own intensity value.
2.2 Physical Constraints
For data acquisition a monostatic laser scanning system is used.
The received energy E r = c-a-w is calculated from amplitude
and width of the received signal approximation. The factor c is
constant and has therefore no influence for our consideration.
Considering an energy balance it depends on the transmitted
energy E t , the distance R to the object surface, and the
incidence angle 3, which is given by the angle between the
transmitter direction and the surface normal vector
where C and C r are constant terms of the transmitter and the
receiver (Kamermann, 1993; Pfeifer et al., 2007). The
atmospheric attenuation along the way from the transmitter to
the object and return to the receiver is describes by
T 2 (R) = e~ 2aR . Let /(cy) entail all other influences like
surface material and local surface geometry. This formula is
valid for objects with larger size than the footprint of the laser
beam. All constant terms may be ignored because at this point
we are interested only on the behaviour of the received intensity.
Knowing the received amplitude and width of the signal a range
corrected intensity is calculated to
I R =C X ■ a ■ w ■ T~ 2 (r) ■ R 2 , (2)
where Q may be any arbitrary constant. This intensity I R
does not dependent on the distance R anymore. For
homogenous regions, cf. the following sections, this formula
delivers value differences less than 5% in comparison with
using a mean distance. This is only valid for data captured at a
single flight path. I R is influenced by the material properties
and the incidence angle.For all points with high planarity we
normalized the measured intensity additionally by
7 = /^/cos(i9) considering the incidence angle. The
illumination direction e t is calculated from the sensor to the
object position. The normal vector of an object surface is
determined by the evaluation of the covariance matrix, cf.
Section 2.3, with respect to the smallest eigenvalue T 3 and its
eigenvector e 3 . With this normalized vectors we calculate the
required divisor by cos(.9) = \e t °e 3 | . These correction steps
remove known influences from the measured intensity.
Therefore the normalized intensity / depends for the used
wavelength on the material properties only. The influence of
speckle effects is neglected.