×

You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Title
CMRT09
Author
Stilla, Uwe

In: Stilla U, Rottensteiner F, Paparoditis N (Eds) CMRT09. IAPRS, Vol. XXXVIII, Part 3/W4 — Paris, France, 3-4 September, 2009
VEHICLE ACTIVITY INDICATION FROM AIRBORNE LIDAR DATA OF URBAN
AREAS BY BINARY SHAPE CLASSIFICATION OF POINT SETS
W. Yao a ' *, S. Hinz b , U. Stilla 3
‘‘Photogrammetry and Remote Sensing, Technische Universitaet Muenchen, Arcisstr.21, 80290 Munich, Germany
b Institute of Photogrammetry and Remote Sensing, Universität Karlsruhe (TH), 76128 Karlsruhe, Germany
KEY WORDS: Airborne LiDAR, Urban areas, Vehicle extraction, Motion indication, Shape analysis
ABSTRACT:
This paper presents a generic scheme to analyze urban traffic via vehicle motion indication from airborne laser scanning (ALS) data.
The scheme comprises two main steps performed progressively — vehicle extraction and motion status classification. The step for
vehicle extraction is intended to detect and delineate single vehicle instances as accurate and complete as possible, while the step for
motion status classification takes advantage of shape artefacts defined for moving vehicle model, to classify the extracted vehicle
point sets based on parameterized boundary features, which are sufficiently good to describe the vehicle shape. To accomplish the
tasks, a hybrid strategy integrating context-guided method with 3-d segmentation based approach is applied for vehicle extraction.
Then, a binary classification method using Lie group based distance is adopted to determine the vehicle motion status. However, the
vehicle velocity cannot be derived at this stage due to unknown true size of vehicle. We illustrate the vehicle motion indication
scheme by two examples of real data and summarize the performance by accessing the results with respect to reference data
manually acquired, through which the feasibility and high potential of airborne LiDAR for urban traffic analysis are verified.
1. INTRODUCTION
Transportation represents a major segment of the economic
activities of modem societies and has been keeping increase
worldwide which leads to adverse impact on our environment
and society, so that the increase of transport safety and
efficiency, as well as the reduction of air and noise pollution are
the main task to solve in the future (Rosenbaum et al., 2008).
The automatic extraction, characterization and monitoring of
traffic using remote sensing platforms is an emerging field of
research. Approaches for vehicle detection and monitoring rely
not only on airborne video but on nearly the whole range of
available sensors; for instance, optical aerial and satellite
sensors, infrared cameras, SAR systems and airborne LiDAR
(Hinz et al., 2008). The principal argument for the utilization of
such sensors is that they complement stationary data collectors
such as induction loops and video cameras mounted on bridges
or traffic lights, in the sense that they deliver not only local data
but also observe the traffic situation over a larger region of the
road network. Finally, the measurements derived from the
various sensors could be fused through the assimilation of
traffic flow models. The broad variety of approaches can be
found, for instance, in compilations by Stilla et al., (2005) and
Hinz et al., (2006).
Nowadays, airborne optical cameras are widely in use for these
tasks(Reinaitz et al., 2006). Yet satellite sensors have also
entered into the resolution range (0.5-2m) required for vehicle
extraction. Sub-metric resolution is even available for SAR data
since the successful launch of TerraSAR-X. The big advantage
of these sensors is the spatial coverage. Thanks to their
relatively short acquisition time and long revisit period, satellite
systems can mainly contribute to the collection of statistical
traffic data for validating specific traffic models. Typical
approaches for vehicle detection in optical satellite images are
described by Jin and Davis, (2007) and Sharma et al., (2006),
and in spacebome SAR images by Meyer et al., (2006) and
Runge et al., (2007). For monitoring major public events,
mobile and flexible systems which are able to gather data about
traffic density and average speed are desirable. Systems based
on medium or large format cameras mounted on airborne
platforms meet the demands of flexibility and mobility. With
them, large areas can be covered (up to several km 2 per frame)
while keeping the spatial resolution high enough to image
sufficient detail. A variety of approaches for automatic tracking
and velocity calculation from airborne cameras have been
developed over the last few decades. These approaches make
use of substructures of vehicles such as the roof and windscreen,
for matching a wire-frame model to the image data (Zhao and
Nevada. 2003).
Despite that LiDAR has a clear edge over optical imagery in
terms of operational conditions, there have been so far few
works conducted in relation to traffic analysis from laser
scanners. On the one hand it is an active sensor that can work
day and night; on the other hand it is range senor that can
capture 3d explicit description of scene and penetrate
volumetric occlusions to some extent. Toth and Grejner-
Brzezinska. (2006) has presented an integrated airborne system
of digital camera and LiDAR for road corridor mapping and
dynamical information acquisition. They addressed a
comprehensive working chain for near real-time extracting
vehicles motion based on fusing the images with LiDAR data.
Another example of applying ALS data for traffic-related
analysis can be found in Yarlagadda et al., (2008), where the
vehicle category is determined by 3-d shape-based
classification.
In this paper, a generic scheme to discover the vehicle motion
solely from airborne LiDAR data is presented. It is based on
two-step strategy, which firstly extracts single vehicles with
contextual model of traffic objects and 3d-segmentation based
classification (3-d object-based classification), and secondly
classifies vehicle entities in view of motion status based on
shape analysis.
2. VEHICLE EXTRACTION
In this step, we need to at first extract various vehicle categories
as complete and accurate as possible, but not considering the
difference among them in terms of dynamical status. To
Corresponding author.