International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004 
  
data became available, usually on all four multiple returns, 
and the laser repetition rate has approached 100 kHz. These 
developments provide an unprecedented point density on the 
ground, which, in turn, helps to accelerate the process of 
moving from simple surface extraction use of LiDAR to 
more sophisticated feature extraction, such as building or 
vehicle extraction (Vosselman and Dijkman, 2001). 
Every indication is that transportation and other agencies will 
be deploying LiDAR systems over transportation corridors at 
an increasing rate in the future — mainly to support 
infrastructure mapping to create accurate surface information 
of highways and areas around highways. Primarily for 
engineering purposes, the road surface must be determined at 
sub-decimeter level accuracy. In general, the vehicles on the 
road represent obstructions to the LiDAR pulses sent to 
reflect off the pavement. Therefore, a substantial amount of 
processing must be devoted to “removing the vehicle 
  
ts 
  
Figure 2. LiDAR intensity data of a 
3. OPTICAL IMAGERY 
Aerial photography was the main technology for airborne 
mapping for decades. As a proven tool, large format aerial 
cameras have provided an enormous amount of spatial data — 
estimates run as high as 95% of geospatial data were 
collected from aerial photography until the late 90's. Digital 
camera systems have been used on airborne platforms for a 
decade, but due to their limited resolution (ground coverage) 
they initially served only remote sensing applications — their 
good radiometric characteristics provide for excellent 
classification performance. As imaging sensor developments 
resulted in larger CCD chips, reaching the 16 Megapixel 
range, the feasibility of building photogrammetric quality 
aerial digital cameras became a reality. Currently, there are 
two main categories of these cameras: (1) medium format 
single sensor frame cameras, such as a 4K by 4K sensor 
based systems, and (2) high-performance large format aerial 
digital camera systems, such as the ADS40 scanner from 
Leica, which is based on linear sensors and multihead camera 
systems, for example, the DMC camera from Z/I Imaging or 
the UltraCam from Vexcel. 
REA 
849 
signals." Rather than removing and discarding the signals, 
however, they can be turned into traffic flow information. 
This way, LiDAR surveys devoted to surface extraction will 
soon be able to provide a valuable byproduct with little or no 
additional effort. It was shown for the first time that civilian 
vehicles could be extracted from LiDAR data with good 
accuracy (Toth et. al., 2003c). As verified below, vehicles 
can be reliably classified into several categories such as cars, 
trucks, and multi-purpose vehicles, based on the pattern of 
the LiDAR returns. With the appropriate LiDAR point 
density, it is expected that vehicle velocities can be estimated 
more reliably in the future. 
Figures 1 and 2 show representative LiDAR data, including 
range and intensity components over a highway segment in 
downtown Toronto, Canada. The data was acquired by a 70 
kHz Optech 30/70 ALTM system. 
  
freeway in Toronto downtown. 
The digital camera systems are mostly different from their 
analog counterpart by the sensor characteristics. The 
complete description of this topic goes beyond the scope of 
this paper; see vendors specifications and, for example, the 
recently published Manual of Photogrammetry. In short, 
CCDs have linear transfer characteristics and, in general, can 
produce much better radiometric data than their analog 
predecessors (scanned imagery). 10-12 bit intensity data are 
quite common, and more importantly, the noise level can be 
as low as the least significant bit. To a great extent, this 
excellent radiometric performance can counterbalance a 
moderate spatial resolution in terms of processing efficiency. 
Earlier digital cameras had a rather low data cycling rate — it 
took seconds to read off the image from the CCD sensors. 
Newer systems, however, can acquire imagery at rates faster 
than one second, and thus multiple overlap can be easily 
obtained at literally no cost — a definite advantage to support 
highly automated processing. Finally, it must be noted that 
the large gap between the parameters of the images acquired 
from various sensor platforms is rapidly shrinking, as the 
resolution of the upcoming satellite systems continues to 
improve from the currently highest 62 cm GSD.