Mobile Mapping System LiDAR-Data Framework, Paul

Download Report

Transcript Mobile Mapping System LiDAR-Data Framework, Paul

Mobile Mapping System LiDAR-Data Framework
Paul Lewis, Conor McElhinney, Bianca Schön, Tim McCarthy
Introduction
Access
Mobile Mapping Systems (MMS) for infrastructural monitoring and mapping are
becoming more prevalent as the availability and affordability of solutions that
generate high accuracy geospatial data has matured. The NCG StratAG XP1
experimental MMS is an example of such a platform (figure 1).
Because of our spatial database approach access to the MMS LiDAR data, using
spatial queries, is much easier. Access queries can be defined programmatically,
allowing for automated processing of the data, or through a simple web interface
where data coverage is plotted and the user can segment the LiDAR using
standard GIS tools for an area they wish to visualise, (figure 5).
Figure 1: StratAG Mobile Mapping System (MMS) XP1
experimental platform.
Figure 5: Selecting LiDAR data using a simple web interface
Navigation
LiDAR
10Hz –
250 Hz
75kHz –
400k Hz
GPS Time
GPS Time
Latitude, Longitude,
Altitude
Latitude, Longitude,
Altitude
Roll, Pitch,
Yaw
Pulse Width,
Reflectance
Table 1: Some of the properties collected by the GPS and LiDAR
equipment from a typical MMS survey.
Processing large volumes of LiDAR data
in a knowledge extraction work-flow is an
active field of research; however, viable
solutions in this context are normally
bespoke,
processing
intensive
operations, for example road-edge
geometry extraction. Typical known
approaches process LiDAR based on a
survey methodology which constrains
the systems to the data that is available
in a single survey.
Figure 2: Over view of the
MMS geospatial data
handling framework model.
Figure 3: Components
discussed in this poster.
In our solution we have developed a
LiDAR data handling framework where
all survey data is searchable. A key
aspect is the ability to globally segment
from this database based on spatially
optimised constraints. In this poster we
present three stages in this process,
which are as follows (figure 3):
1. STORAGE: Workflow used to upload,
store and index very large volumes of
LiDAR and Navigation data.
2. ACCESS: Workflows and interfaces for
defining optimised spatial queries for
LiDAR data retrieval.
3. PROCESSING: Results and visualisations
from processing spatially segmented
LiDAR through knowledge extraction
algorithms.
Storage
Our storage solution is built upon
a PostgreSQL database with
PostGIS spatial extensions. Data
input and index generation is
being handled through the
pg_bulkload high speed data
loading utility, (figure 4). It is
currently being tested but early
results are significant, (Table 2).
More
comprehensive
and
comparative test results are
currently being compiled.
Figure 4: pg_bulkload published results
for high speed data loading.
Upload Procedure
Time hr:min Points
pg_bulkload;
20:53
2 geometries & 2 GIST indexes
SQL Copy;
18.48
2 geometries & 2 GIST indexes
77 m
25 m
Table 2: Preliminary Post-Survey Lidar-data
upload timing results to DB.
We performed 27 different spatial
queries on cached data by varying the
length (15m, 30m, 60m) the width (1m,
10m, 20m) and the height (1m, 2m, 5m,
no limit) of the queries, and executed
each query 9 times. An average was
taken, example times are shown in
Figure 6. These demonstrate that 3D
queries were faster than 2D queries.
We believe this is based on the limiting
factor for query execution which we
found to be the number of points being
returned. The query execution time was
0.0000935s per point with a standard
deviation of 0.0000167.
70
60
2D 10m (width)
3D 10m*5m (width*height)
3D 10m*2m (width*height)
2D 20m (width)
3D 20m*5m (width*height)
3D 20m*2m (width*height)
50
Time (s)
The spatial data properties being collected on the XP1 MMS platform are
shown in table 1 and are used in the development and testing of our storage,
access and processing solution. This poster presents work that has been
undertaken to optimally handle the large volumes of spatial data, specifically
the MMSs navigation and LiDAR data. This work is defined as part of a
framework (figure 2) for MMS spatial data handling that is being developed
upon the open source spatial database technology PostGIS. Our results
analyse the I/O challenges with handling these data and discuss how this
solution is optimal in this context.
40
30
20
10
15
20
25
30
35
40
Length (m)
45
50
55
60
Figure 6: Timing information for spatial
queries of different dimensions.
Processing
We are developing algorithms to
automatically extract a range of
features from LiDAR data, Figure 7.
Using the spatial database we can
select the dimensions and the
geographical area that suits the
algorithm.
For
instance,
when
extracting the road we select points +10m from the navigation point and +2m below the vehicle.
With our partners in ITC
we
are
developing
automated algorithms for
segmenting trees, man
made poles, road edges
and the road surface
from LiDAR scans. In
Figure 8, the results of
this processing on a
LiDAR
segment
are
displayed.
Figure 7: Tree extracted
from LiDAR data.
* Road Edge
* Tree
* Man-made
Pole
Figure 8: Processed LiDAR showing trees, poles
and the road edge.
Conclusions and Future Work
This research, through preliminary testing and profiling, has shown that it is feasible
for a PostgreSQL database with PostGIS spatial extensions to be used to efficiently
store very large volumes of MMS LiDAR data. While significant issues exist with data
preparation and input to any LiDAR framework, we have shown how these can be
minimised through optimised processes using pg_bulkload. The significant
advantage from this framework approach is the ability to dynamically access any
LiDAR data from the database for any pocess. This LiDAR segmentation is done
spatially, either through a visual interface or in a programmatically controlled context,
and can optimise LiDAR data sets for distinct requirements as has been shown in the
case of the road geometry extraction algorithm.
Future work involves moving this framework onto a dedicated server hardware
platform from which further test results and analysis can be performed. But also the
framework will be extended to incorporate the large volumes of imagery that will also
be captured from the MMS.
Research presented in this poster was funded by a Strategic Research Cluster Grant (07/SRC/I1168) by Science Foundation Ireland under the National Development Plan and by the ERA-NET SR01 projects. The authors gratefully acknowledge this support.