20-06-2014, 10:33 AM
A novel geo-localisation method using GPS, 3D-GIS and Laser scanner for intelligent vehicle navigation in urban areas
[attachment=65087]
Abstract
This paper tackles the problem of vehicle’s geolocalisation
in urban areas. For this purpose, Global
Positioning System (GPS) receiver is the main sensor. But the
use of GPS alone is not sufficient in many urban environments.
GPS has so to be helped with dead-reckoned sensors, map data,
cameras, range finder … In this paper, we propose a novel
approach to compute observation of the absolute pose of the
vehicle to back up GPS and to compensate the drift of deadreckoned
sensors. This approach uses a new source of
information which is a 3D city model i.e. 3D city model of the
environment of vehicle evolution. This 3D city model is
managed in real-time by a 3D Geographical Information
System (3D-GIS). The pose’s observation is constructed by
using an on–board horizontal laser scanner which provides a
set of distances. This set of distances (laser scan data) is
matched with depth information (virtual laser scan data),
provided by 3D-GIS, using Iterative Closest Point algorithm
(ICP).
Experimental results performed using real data illustrate the
performances of the proposed approach
INTRODUCTION
Real-time positioning system is a very important module
for intelligent vehicle and/or transport systems and
applications [1]. Three dimensional city models can be very
important sources information for navigation applications.
As a matter of fact, real-time positioning on 3D city model
allows to accurately localizing the vehicle on its evolution
environment. It can be also very useful for obstacle detection
[2]. A 3D city model appears like the natural evolution of
the 2D digital map which is actually widely used in
navigation system, Advanced Driver Assistance System…
Outdoor positioning systems often rely on GPS [3],
because of its affordability and convenience. However, GPS
suffers from satellite masks occurring in urban
environments, under bridges, tunnels or in forests. GPS
appears then as an intermittently-available positioning
3D CITY MODEL AND 3D-GIS
More and more applications require 3D city models of the
buildings, cities, landscapes, in addition to 2D digital maps
commonly managed by Geographical Information Systems
(GIS), aims to model the real world as accurately as
possible. These 3D city models, also called cartographic 3D
model or virtual 3D model in the literature, actually can be
automatically generated using aerial images, 2D digital map,
laser profiler data, land registry and so on.
The German GEIST project shows that such 3D city
models enable a better understanding of spatial relationships
and are therefore ideal for demonstration and presentation
purposes. A 3D city model of the environment can be a
support tool to check the coherence of an architectural
project or to detect possible anomalies. Applications and
markets of 3D city models vary a lot from urban and master
planning to car navigation.
Thanks to such applications, 3D city models meet a rapid
expansion [7] [8]. For example, all the major cities in Japan
have been covered since 2002 and are updated every six
months. The French national geographic institute (IGN)
which provides 2D maps aims to create the geographical 3D
map of France for 2010 in its Bati3D project. The
indisputable appeal of 3D city models leads us to propose an
utilisation of these ones in the intelligent transportations
domain. We propose to construct a 3D
GEO-LOCALIZATION METHOD
After a calibration step, we have identified the elevation
of the on-board laser scanner, fixed on the top of the vehicle,
in the 3D city model. Supposing that the vehicle is running
in a plane surface, we suppose that is no pitch and no roll. So
the elevation (z) of the laser scanner with a predicted pose
(x, y,θ) permit to extract a depth image from the 3D city
model managed by 3D-GIS and depth file which is attached
to. In the same horizontal plane, we can consider that
distances provided by the laser scanner (laser scan data) is
comparable to distances observed by the virtual camera
(virtual laser scan data) provided by the depth file. At this
step, the task is to match a pair of scans: virtual laser scan
data with the laser scan data. To this task, we use the ICP
algorithm [9]. To compute the pose, the ICP output
transformation R and T are finally used with the kinematic
of a unicycle [10]. The synoptic of the proposed method is
described in Fig.3.
3D coordinates’ determination
The 3D-GIS returns the virtual image and a text file
containing the depth information which is the distance in
meter between the virtual camera and the 3D points
represented by the pixels of the virtual image. We have to
compute the 3D coordinates of the points that are projected
in the virtual image plane.
Consider p a pixel of the virtual image. Let denote (u, v)
the coordinates of this pixel p in the image plane. Let denote
P’ the corresponding 3D point that is projected at p in the
virtual image [12].
ICP Algorithm & Pose Estimation
The ICP (Iterative Closest Point by Besl & Mckay) [13]
algorithm provides a simple and effaces range-image
registration of 3D shapes techniques (Fig.7). Range–image
registration, also called scan alignment or unprocessed data
correlation, is the process of aligning an observed set of (2-D
or 3-D) points with a reference point set.
Objects are represented by free-form curves, and a freeform
curve is available in the form of a set of chained points.
Therefore the main concept of ICP is based on matching the
scanned data to an existing map in the form of a curve that
requires only a procedure to search for the closest point on a
geometric entity to a given point, using a nearest neighbour
algorithm. Secondly a least-squares technique is used in
accordance with the complex level of objects’ geometric
shape, give the appropriate transformation (rotation “R”
and/or translation “T”) parameters from the point
correspondences. Thus iteratively applying this procedure,
the algorithm yields a better and better estimate and
gradually minimizes mean-square distance metric between
curves in the two sets.
REAL EXPERIMENTATION
Experimentation has been carried out around LORIA Lab
with embedded sensors on the CyCab (which is a robosoft
product (www.robosoft.fr)): the GPS RTK Thales Sagitta 02
system, the laser SICK LMS 291 (with a viewing angle of
180° and the range is 80m) and the gyroscope (give yaw of
vehicle relative to the magnetic north axis) figure
CONCLUSION
In this paper, we have presented a method which takes
advantage of a 3D city model combined with Laser scanner
to compute a geo-pose for navigation applications in urban
area. The proposed method has been tested and validated
with real data. Obtained experimentation results are
promising and it illustrates the benefits of a 3D city model
for geo-localisation in urban areas.
The proposed method can be a part of a multi-sensor data
fusion framework to take advantage of information
redundancy. It’s noticed that the result of the proposed
method can be affected by pitch and roll angle if the vehicle
[attachment=65087]
Abstract
This paper tackles the problem of vehicle’s geolocalisation
in urban areas. For this purpose, Global
Positioning System (GPS) receiver is the main sensor. But the
use of GPS alone is not sufficient in many urban environments.
GPS has so to be helped with dead-reckoned sensors, map data,
cameras, range finder … In this paper, we propose a novel
approach to compute observation of the absolute pose of the
vehicle to back up GPS and to compensate the drift of deadreckoned
sensors. This approach uses a new source of
information which is a 3D city model i.e. 3D city model of the
environment of vehicle evolution. This 3D city model is
managed in real-time by a 3D Geographical Information
System (3D-GIS). The pose’s observation is constructed by
using an on–board horizontal laser scanner which provides a
set of distances. This set of distances (laser scan data) is
matched with depth information (virtual laser scan data),
provided by 3D-GIS, using Iterative Closest Point algorithm
(ICP).
Experimental results performed using real data illustrate the
performances of the proposed approach
INTRODUCTION
Real-time positioning system is a very important module
for intelligent vehicle and/or transport systems and
applications [1]. Three dimensional city models can be very
important sources information for navigation applications.
As a matter of fact, real-time positioning on 3D city model
allows to accurately localizing the vehicle on its evolution
environment. It can be also very useful for obstacle detection
[2]. A 3D city model appears like the natural evolution of
the 2D digital map which is actually widely used in
navigation system, Advanced Driver Assistance System…
Outdoor positioning systems often rely on GPS [3],
because of its affordability and convenience. However, GPS
suffers from satellite masks occurring in urban
environments, under bridges, tunnels or in forests. GPS
appears then as an intermittently-available positioning
3D CITY MODEL AND 3D-GIS
More and more applications require 3D city models of the
buildings, cities, landscapes, in addition to 2D digital maps
commonly managed by Geographical Information Systems
(GIS), aims to model the real world as accurately as
possible. These 3D city models, also called cartographic 3D
model or virtual 3D model in the literature, actually can be
automatically generated using aerial images, 2D digital map,
laser profiler data, land registry and so on.
The German GEIST project shows that such 3D city
models enable a better understanding of spatial relationships
and are therefore ideal for demonstration and presentation
purposes. A 3D city model of the environment can be a
support tool to check the coherence of an architectural
project or to detect possible anomalies. Applications and
markets of 3D city models vary a lot from urban and master
planning to car navigation.
Thanks to such applications, 3D city models meet a rapid
expansion [7] [8]. For example, all the major cities in Japan
have been covered since 2002 and are updated every six
months. The French national geographic institute (IGN)
which provides 2D maps aims to create the geographical 3D
map of France for 2010 in its Bati3D project. The
indisputable appeal of 3D city models leads us to propose an
utilisation of these ones in the intelligent transportations
domain. We propose to construct a 3D
GEO-LOCALIZATION METHOD
After a calibration step, we have identified the elevation
of the on-board laser scanner, fixed on the top of the vehicle,
in the 3D city model. Supposing that the vehicle is running
in a plane surface, we suppose that is no pitch and no roll. So
the elevation (z) of the laser scanner with a predicted pose
(x, y,θ) permit to extract a depth image from the 3D city
model managed by 3D-GIS and depth file which is attached
to. In the same horizontal plane, we can consider that
distances provided by the laser scanner (laser scan data) is
comparable to distances observed by the virtual camera
(virtual laser scan data) provided by the depth file. At this
step, the task is to match a pair of scans: virtual laser scan
data with the laser scan data. To this task, we use the ICP
algorithm [9]. To compute the pose, the ICP output
transformation R and T are finally used with the kinematic
of a unicycle [10]. The synoptic of the proposed method is
described in Fig.3.
3D coordinates’ determination
The 3D-GIS returns the virtual image and a text file
containing the depth information which is the distance in
meter between the virtual camera and the 3D points
represented by the pixels of the virtual image. We have to
compute the 3D coordinates of the points that are projected
in the virtual image plane.
Consider p a pixel of the virtual image. Let denote (u, v)
the coordinates of this pixel p in the image plane. Let denote
P’ the corresponding 3D point that is projected at p in the
virtual image [12].
ICP Algorithm & Pose Estimation
The ICP (Iterative Closest Point by Besl & Mckay) [13]
algorithm provides a simple and effaces range-image
registration of 3D shapes techniques (Fig.7). Range–image
registration, also called scan alignment or unprocessed data
correlation, is the process of aligning an observed set of (2-D
or 3-D) points with a reference point set.
Objects are represented by free-form curves, and a freeform
curve is available in the form of a set of chained points.
Therefore the main concept of ICP is based on matching the
scanned data to an existing map in the form of a curve that
requires only a procedure to search for the closest point on a
geometric entity to a given point, using a nearest neighbour
algorithm. Secondly a least-squares technique is used in
accordance with the complex level of objects’ geometric
shape, give the appropriate transformation (rotation “R”
and/or translation “T”) parameters from the point
correspondences. Thus iteratively applying this procedure,
the algorithm yields a better and better estimate and
gradually minimizes mean-square distance metric between
curves in the two sets.
REAL EXPERIMENTATION
Experimentation has been carried out around LORIA Lab
with embedded sensors on the CyCab (which is a robosoft
product (www.robosoft.fr)): the GPS RTK Thales Sagitta 02
system, the laser SICK LMS 291 (with a viewing angle of
180° and the range is 80m) and the gyroscope (give yaw of
vehicle relative to the magnetic north axis) figure
CONCLUSION
In this paper, we have presented a method which takes
advantage of a 3D city model combined with Laser scanner
to compute a geo-pose for navigation applications in urban
area. The proposed method has been tested and validated
with real data. Obtained experimentation results are
promising and it illustrates the benefits of a 3D city model
for geo-localisation in urban areas.
The proposed method can be a part of a multi-sensor data
fusion framework to take advantage of information
redundancy. It’s noticed that the result of the proposed
method can be affected by pitch and roll angle if the vehicle