Istanbul 2004
b) Model
e of detected
it can hardly
ve covers all
1 that such a
«es it clearly
reasons the
more global
| are relaxed
y of features
'ometric and
ted and, in
of vehicles,
ire. ("ladder-
According to
earching for
'ed- by (i)
edges with
(11) frequent
| edges with
e, (it) high
ine and (iv)
> symmetry
and queue
nd queues is
are further
] axis. After
icles. Such
obs that can
last step, all
odel but not
les detected
; recovered
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
3. THERMAL INFRARED DATA
3,1 Related work
Infrared sensors open the possibility of night vision and sensing
the traffic situation by day and night. Vehicle movements can
be captured by commercial IR-Cameras with a frame rate of 50
Hz (standard video out) An experimental airborne system
including GPS and INS was built up within the LUMOS
Project. It was shown that a tracking of moving objects
exploiting the georefenced image sequence and map data is
possible [Sujew & Ernst, 2003].
The activity of vehicles is not restricted to movement features
alone. Stationary cars may still have the engine idling. They
may be waiting at a traffic light or in a traffic jam. This is
typical for rush hour situations in cities. Regarding e.g.
pollution such cars have to be counted as active. Furthermore,
for traffic management it might be helpful to distinguish cars
that halt only for a short time - e.g. for loading, in a drive
through service or just quickly picking things up - from those
who are parked for hours or days. Temperature is an important
feature for such recognition tasks. It is known that temperature
can be remotely sensed by thermal cameras sensitive in the 3-
5um or 8-12pum spectral band [Wolffe & Zissis, 1985].
f g
Figure 3. a) Example of a thermal IR image of an urban area.
Enlargement of white frames see c-g. Parking cars
appearing as dark spots which are grouped in rows
along the street, b) Interest regions superimposed by
a map on the IR-image, c) hot spot of a chimney, d)
hot spot of a street light, e)-f) parking cars, one
arrived recently arrived, g) car moved away
32 Detection of single vehicles
Stationary passive cars appear as dark stationary spots, because
they are colder than the surrounding. They will usually occur
along the margin of roads or in parking lots, and they will be
grouped into rows. Fig. 3a shows such a situation. Active
vehicles will appear as moving bright spots on the roads. A
stationary bright spot within a row of stationary dark spots can
be interpreted as a car that is still warm (has been moved short
time before)(Fig. 3e,f) or as warm spot on the bare concrete
giving the hint, that there has been a vehicle short time before,
that moved away (Fig. 3g). Rows of bright spots in the roads
are probably cars, which are waiting at a traffic light or in a
traffic jam. Thus the percentage of bright versus dark spots
gives a good estimation of the activity of cars in the scene. Fig.
975
3a shows a typical example of a thermal image of an urban site
with rather low car activity. While some buildings and cars
have good contrast to the background, roads do not necessarily
have clear margins in this kind of imagery.
3.3 Exploiting context
Cars tend to be placed in equidistantly spaced rows along the
margin of roads. This criterion allows discrimination from other
spot-shaped objects. Grouping of such spots into rows of
arbitrary length is a generic operation. Fig. 4a shows a section
of an IR image containing a row of cars.
All detected warm and cold spots in this section are displayed
in Fig. 4b. Spots caused by cars constitute only a subset of
these. Fig. 4c shows those spots, that are sufficiently close to a
road margin. The grouping starts only from spots which exceed
a minimal mass. The grouping direction is constrained by the
road margin. Only spots fitting into the straight and
equidistantly spaced row model are grouped.
Still, there may be several alternatives of grouping, e.g. if two
spots are close to one another in a location consistent with the
model (see Fig. 4c, most right member of the row). Among the
alternatives one group is selected based on an assessment,
which is calculated from the number of spots, total mass,
regularity in spacing and straightness and consistency in
orientation of the spots. Fig. 4d shows the best group containing
seven spots.
€ | d
Figure 4. The benefit of grouping: a) A section of an IR-image;
b) all spots constructed in that region; c) spots in the
interest region given by fusion with the map; d) car-
spots remaining consistent with the row model after
grouping.
4. SAR DATA
Active synthetic aperture radar (SAR) sensors are independent
of the time of day. The large signal wavelength in the order of
centimetres provides near insensitivity to weather conditions.
The increasing resolution of SAR data offers the possibility to
utilize this data for detailed scene interpretations in urban areas.
