rt B2. Istanbul 2004
T CENTER OFFSET
ic Diagram
he hotspot from the
s been enhanced for
values. At the flying
e fire pit imaged here
tspot from the fire in
f the blooming effect
2 shows the effect of
can see some of the
act centre of the un-
he hot spot occurs, a
1Se some error in the
ng the location of the
ot, it may be possible
d also to potentially
'ctions and low level
wn in the following
cd in Figure 3. Note
otspots are black. To
fire hot spots from
as been defined, as
ghlighting
Spots
ige for clarity)
mal Images
iny features that are
4 outlines this step.
y pixels and object
om the background
ting a feature mask.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
Threshold
—————-
Background Adjusted Image Feature Mask
Original
Feature Mask Result
Figure 4: Image Processing for Feature Identification
The feature mask is used to select isolate particular features
from the original image. The background adjustment process
(not described here), reduces noise to simplify the feature
extraction process, but tends to eliminate the small hot spot
peak temperature characteristic, so the original image is used
for peak pixel intensity determination. The desired features are
extracted from this final resulting image. Several
characteristics of the features are extracted, including the center
of gravity, area, maximum intensity pixel, and the
corresponding coordinate of the maximum intensity pixel.
These feature characteristics are required for the second step of
the process, outlined next.
2.2 Hot Spot Feature Track Identification
By using a low threshold in the previous step to isolate any
features of above average temperature, several undesired
features are also extracted with the hot spots. So instead we
turn to the temporal aspect of the image and feature sequence.
Fire is extremely obvious, by the very high pixel intensity,
when the thermal imager directly views it. As stated previously,
the direct viewing path may be obstructed, such that a hot spot
may only show up in one or two few images at a time. For
various reasons, a longer feature track is desirable for
accurately identifying, tracking and locating any hot spots.
Only the basics required to specifically identify a hot spot are
presented here.
Every feature previously extracted is cross-referenced or
tracked from frame to frame, creating a feature track. The
maximum observed pixel intensity is updated for that feature
track until no more features are added to that track for several
frames. Once the feature track is considered complete, the
maximum pixel intensity is evaluated as to whether that track
was ever hot enough to be considered a fire or not.
Figure 5 is an actual intensity profile observed during testing.
For a majority of the feature track, the pixel intensity stays
fairly low, and varies quite substantially. It peaks above 250
for one pixel towards the end of the feature track. This peak
value of the track, and in some respects the range of values, are
what identify this feature track as the track of a hot spot.
CA
£ 150 v
= * 4
= T PSP P * *
É 300 : fe
c * *
$m 1, 4t * *.
5 —*- * a
A ++ » x
= 100 | 4
X I i
à "I que
25900 — 25820 25040 25360 25880 26000 26020
Figure 5: Example Hot Spot Feature Track Intensity Profile
Once the hot spot feature track is identified, then the motion of
the hot spot is tracked over different images. Motion estimation
is the determination of spatial change over temporal changes.
The first step in tracking the hot spots from frame to frame is to
develop an estimate of the potential or effective pixel motion
that could occur. This involves both the camera characteristics
and also the overall motion and flight characteristics. These
include the effective flying height, flight speed, crab angle, and
attitude variation of the system, as well as the effective field of
view of the camera. On a frame by frame basis, there is a direct
correspondence between the change in roll and change in image
x coordinate (dx), as there is with the change in pitch and
change in image y coordinate (dy). The crab angle, or
difference between actual motion vector and the body
orientation will have a slight affect on both x and y, adding
additional variation to both dx and dy. The majority of the
image flow will occur from the position changes of the aircraft
instead of the attitude. The change in image y coordinate (dy)
is related to the forward motion of the aircraft, and dx is related
to any side to side motion. Finally, the actual pixel motion due
to attitude variations is independent and unrelated to height, but
pixel motion due to speed is directly related to the height, due
to the change in scale with height.
In a remote sensing application, the attitude of the aircraft is
generally controlled very tightly such that the variations are
minimal, although a range will be considered here. Table 1 lists
the pixel motion range estimates due to attitude variations, and
Figure 5 shows the pixel motion range estimates due to height
and flight speed. Figure 6 shows an example of an entire hot
spot feature track, as it would be seen moving across the image.
Once a hot spot is tracked through the image sequence, its
location can be determined, as explained in the following
section.
Angle X Y
Type Degrees Pixel Variation
Roll: 0.5 2.61 0.00
Pitch: 0.5 0.00 2.61
Azimuth: 0.5 1.38 1.38
Total Pixels: 399 3.99
Table 1: Pixel Motion Estimate Due to Attitude Variations
