International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004 
  
  
  
  
For the framework of the NCRST-F research, the focus was 
on using 4K by 4K cameras, as this category is frequently 
used as a companion digital camera for LiDAR systems. In 
addition, the availability as well as the cost prohibited the use 
   
4. TRAFFIC FLOW 
Vehicle traffic over the transportation infrastructure affects 
almost every facet of our life and has a primary impact on the 
economy. As vehicle traffic continues to grow while the 
resources to increase the road capacity are limited, the only 
answer to keep up with the ever-increasing traffic is better 
management. This, in turn, depends on the availability of 
better traffic data, i.e., timely information on almost every 
vehicle traveling on the transportation network. 
Traffic flow is a generic term used to describe vehicle 
movement and volume over a transportation network; two of 
the most important traffic measures produced by state DOTs 
and other transportation agencies around the world are 
AADT and VMT (Pline ed., 1992). Average annual daily 
traffic (AADT) is produced to represent the vehicle flow over 
a highway segment on an average day of the year. Vehicle 
miles traveled (VMT) indicates travel over the entire 
highway system and is used to indicate mobility patterns and 
travel trends. VMT is also used as an indicator for allocation 
of highway resources. Flow data are generally obtained by 
ground-based equipment, such as loop detectors or road tubes, 
which are fixed to a location and are deployable as needed. 
In the latter case, the sample data are collected from road 
tubes placed in the traveled portion of the road, disrupting 
traffic and endangering the crews when placing or collecting 
the tubes. Using satellites and air-based platforms, the 
survey/control crews can cover large areas, access remote 
highways, and carry sensors that can collect data from safe 
and non-disruptive off-the-road locations. The imagery 
collects "snapshots" of traffic over large areas at an instant of 
time or a sequence of snapshots over smaller areas, whereas 
traditional data collection observes vehicles at a point on the 
highway over much longer time intervals (McCord et. al., 
2003). 
Short-term flow parameters, including daily or hourly 
parameters broken down to vehicle categories are also of 
high interest, but can be acquired only with a limited spatial 
extent with traditional techniques, as the density of ground- 
based sensors cannot be increased infinitely. The use of 
airborne imagery, however, offers an on-demand deployable 
tool with excellent temporal and spatial resolutions that can 
easily provide for sizeable area coverage at fast sampling 
rates. Therefore, the primary objective of this investigation 
was to research the feasibility of obtaining short-term flow 
data by using remote sensing technologies. 
Figure 3. 4K by 4K digital camera ortho image of a freeway in Toronto downtown. 
850 
of the high-end digital camera systems until recently. A 
representative 4K by 4K image, acquired by the DDS digital 
camera system simultaneously with the LiDAR data, is 
shown in Figure 3. 
5. VEHICLE EXTRACTION, GROUPING AND 
TRACKING 
The investigation of extracting flow data from airborne 
remote sensing has been carried out using two different 
sensor configurations: (1) to monitor the feasibility of 
obtaining traffic flow data over road segments, LiDAR data 
were acquired from several test flights, and in one case, 4K 
by 4K imagery was simultaneously captured and then used as 
the ground truth for the LiDAR data, and (2) to assess the 
potential of flow monitoring at intersections, a 4K by 4K 
digital camera and a standard video system were installed on 
a helicopter platform and several test flights were conducted. 
5.1 Vehicle Extraction and Grouping From LiDAR Data 
In the first step of the data processing, the input LiDAR data 
are filtered to reduce the point cloud to the road area. The 
approximate location of the road geometry is usually 
available from CAD or GIS databases, maintained by 
transportation agencies. Since the accuracy of the road 
location information is limited, or only the centerline data are 
available, it is mandatory to perform a road-matching step. 
During this process, the edge lines of the road are tracked 
from the LiDAR data. Once the road surface area has been 
identified, the vehicles can be extracted by a simple 
thresholding. During this process the road surface modeled 
and the surface normal are considered, thus, the resulting 
vehicle points are reduced to the height with respect to the 
modeled road surface (road invariant description of the 
vehicles). 
The point cloud of a vehicle can contain a varying number of 
points, mainly depending on the laser point density and the 
relative speed between the vehicle and LiDAR sensor. The 
effect of the latter factor is more important and demands a 
parameterization of points that can, at least, partially reduce 
the effect of vehicle shortening and elongation (compare 
vehicles in Figures 1-2 to Figure 3). The selection of 
parameters has a major impact on the classification process, 
namely, how reliable can the different vehicle groups be 
separated. The basic model was formed from six parameters, 
including the length and width of the vehicle and four height 
values, representing an average height over four equal 
segments of the vehicle. Figure 4 shows the interpretation of 
the basic parameters. 
  
Interne 
  
In : 
ave 
der 
vol 
phy 
An 
sig 
dat 
sub 
PC 
froi 
resi 
can 
  
The 
cat 
truc 
cla: 
sho 
45