156 PDF
156 PDF
156 PDF
ICWG V/I
ABSTRACT:
This paper introduces a complex, low-cost road pavement measurement system designed primarily for pothole and crack detection.
The onboard system is composed of a GPS/INS navigation unit, an image acquisition module, and a photogrammetric and image-
processing subsystem. Due to the use of structured light and the availability of accurate navigation data, the 3D coordinates of the
road points, in the form of cross profiles, as well as the size and location of the potholes, can be derived from the images at cm-level
relative accuracy. For the road classification the International Roughness Index (IRI) can be derived. This paper also discusses the
potential future developments of the system; it reports on the initial investigations of applying single camera with laser line
projection which can significantly simplify the overall procedure.
In the well-motorized countries the transportation authorities Our basic idea was to create a road pavement detection system,
spend significant amounts of money to road constructions and which uses optical sensors for measuring the longitudinal and
road pavement maintenance. The roads are aging; the cross sections of the roads and the pavement anomalies
maintenance works have to keep pace with the new (potholes, wide cracks). Figure 1 shows how the system
constructions. The traffic delays are mostly caused by road components are linked to each other. The sensor component
closures, this topic is always highlighted in the news. Therefore, consists of two cameras (Sony XCD-SX910) synchronized by
the optimized scheduling became extremely important, the external trigger and laser projectors, the location component
authorities tend to turn to new technologies in order to map the contains an integrated GPS/INS (Global Positioning System /
actual condition of the road pavement and decide upon the Inertial Navigation System) unit (Crossbow NAV420CA).
priorities of the scheduled maintenance works.
The cameras are capturing downward images about the
Since road engineering is one of the most extended engineering pavement at a rate of 5 fps. 20 individual laser projectors
field, many “traditional” road surveying methods used for provide the equally spaced marker points located perpendicular
construction and maintenance planning. As most of the to the travelling direction on the surface. Therefore, the 3D
maintenance works are scheduled based on experiences, the coordinates of the points can be computed by evaluating
road conditions are often classified by visual observations. The equations of geometric constraints.
number of potholes, the length and type of the cracks, and lane
grooves are clear indicators of the overall road condition. The georeferenced position and orientation of the camera pair
are provided by the integrated GPS/INS system. The
However, there are a few autonomous (or at least semi- NAV420CA system consists of a navigation grade GPS receiver,
autonomous) systems in use, e.g. the Swedish RST an entry-level IMU (Inertial Measurement Unit), and a built-in
(http://www.opq.se) or the Mandli Roadview Kalman-filter. The accuracy of the positioning by the GPS
(http://www.mandli.com): both systems are able to measure the module is about 3 m CEP, whereas the random walk of the IMU
cross- and longitudinal profiles of the roads; moreover, Mandli is about 4.5°/hr1/2 (Barsi et al. 2005).
is collecting digital images about the road surface.
The cameras have 1280 by 960 pixel resolution which results in
This paper discusses the structure and results of the original 3 mm ground pixel size, which enable not only the detailed
system, the potential of the entry-level IMUs (inertial surface description but the scar and crack detection from the
measurement unit) in mobile mapping and introduces the latest images, too. The camera location assures the full lane-width
concept with the initial test results. visibility, covering 3.5 m wide area. The cameras are linked by
FireWire to the laptop mounted in the cab (Figure 2).
* Corresponding author.
897
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008
Camera 1
Camera 2
GPS Unit
Power
Control
Data
Figure 2. The instrument platform mounted on the pickup and the controlling environment
3. INITIAL RESULTS
898
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008
In all cases the objects have been moved towards the camera,
perpendicularly to the projected line. Sequences of camera
images were the inputs of the overall object reconstruction
procedure. In all images the red laser stripe was extracted by
filtering and binary image morphology, and then the y image
coordinates of the laserline were computed.
899
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008
6. DERIVING THE INTERNATIONAL ROUGHNESS The IRI calculation is done at the highest resolution, but of
INDEX (IRI) course the values can be aggregated arbitrary.
The IRI measurement has m/km (sometimes mm/m) or in/mi An additional experiment was also performed: the potential of
units. The determination of IRI can be easily done from the vertical acceleration directly in IRI calculation i.e. how can IRI
intersected surface points; only the longitudinal profile has to be derived from the single vertical acceleration measurements.
be used. The most relevant place of road roughness is exactly To test this hypothesis, regular gridded vertical acceleration
under the tires, therefore the four (2 left and 2 right) markers in values were interpolated, then they were integrated into vertical
line with the tires have been selected for IRI calculations. The movements, and at the end the IRI-calculations were performed
IRI values are calculated for all four points, samples taken in as (Sayers & Karamihas 1998) refers. The IRI-values are
every 80 cm (at a speed of 4 m/s [15 km/h]) along the road, then calculated for 2 m long segments. Since only a single IMU was
their average are ordered to the analyzed road segment. applied in the vehicle, no smoothing (e.g. averaging) was
needed (compared to the previously mentioned “optical” IRI
A further software module has been developed, which produces method).
IRI files (reports) and IRI maps. In Figure 7 the effect of the
recently finished road reconstruction on the bridge and the lack
of it on the east (Pest) side can be clearly seen.
S1 S2 S3 S4 S5
Cat. 1 2606 1340 1475 518 979
Cat. 2 4456 1797 1968 978 842
Cat. 3 2408 1269 707 753 402
Cat. 4 1228 737 362 466 223
Cat. 5 1785 1143 401 773 342
Cat. 6 2266 1615 336 820 946
Length 14752 7904 5251 4312 3737
Mean 9.32 10.00 7.70 11.54 8.28
IRI
IRI SD 7.28 7.51 6.72 8.52 7.19
900
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008
Considering different pavement types the IRI-diagrams can be Figure 8 illustrates the IRI diagrams for the investigated
compared (Figure 8). pavement types, calculated from the raw vertical acceleration
measurements.
The standardized IRI values for airport runways and
superhighways are 0.5-2.0, 1.5-3.5 for new pavement, 2.5-5.5 The first three pavement types are the smooth surfaces; they
for older pavement, 4.0-11.0 for damaged pavement and 8.0- have an average IRI value of about 0.1 with a relatively small
20.0 m/km for rough unpaved roads. Comparing the IRI values (< 0.05) standard deviation. Compared to this, the last two
of Figure 7, relative good fitting can be noticed: the renewed pavements are worse as the observed higher IRI values (0.3 and
pavement on the bridge has an average IRI value of 4 m/km, 0.6) with larger standard deviation (> 0.05) clearly indicate it.
whilst the older road segment on the east river side has about 16
m/km. Especially this latter value points out that our system Although there are numerically greater differences between our
requires a calibration run to get the mathematical formula to mean IRI values and the standardized ones, the tendency is
assign the IRI observations to the standardized evaluation. obvious: the worse road gets higher IRI measures. This solution
also cannot avoid the calibration procedure.
Figure 9. IRI road status in Budapest derived by the last measurement campaign
The most important statistics of the IRI-values are the following: the marker points. This CAD-based module not only computes
the profiles from the given points, but generates a road surface
model that represents the pavement condition and can be used
Pavement type Mean Std for maintenance assessment, scheduling, and planning.
deviation
Stone mosaic 1 0.1041 0.0195 The investigations and tests described in this paper proved that
Stone mosaic 2 0.1095 0.0164 the proposed single-camera detection method assures robust
Stone mosaic 3 0.1416 0.0317 solution for road surface generation. Applying a single imaging
Small cubes 0.6795 0.1586 unit simplifies the georeferencing and avoids the complicated
Large cubes 0.3543 0.0775 calculation and calibration of two cameras which have to be
synchronized. Using line projection instead of laser point array
Table 2. Basic statistics of IRI on different pavement types enables measuring all point heights along the profile, therefore
(derived from Figure 8 diagrams) the surface model resolution depends only on the horizontal
resolution of the camera. This concept also has further potential
in development, e.g. applying multiple projected lines in order
7. CONCLUSIONS to allow higher measurement speed, and also leaves open the
possibility of using the camera images for additional purposes,
In this paper we presented a concept study and our initial results such as crack detection, which is a critical issue regarding road
of the developed pavement detection system. Digital cameras condition detection.
capture the images of the road pavement which is lit by
structured light. The exterior orientation parameters of the The described mobile mapping system and the provided surface
images are provided by a GPS/INS navigation system, which data can be used for deriving the IRI for the particular road
enables the pavement surface generation at good accuracy. segments, which is used by the transportation authorities for
Applying a special diffuse illumination, the small anomalies of road pavement classification.
the pavement, such as scars and potholes, can also be detected.
In the context of this effort we developed a road profile and The next step in development is creating the engineering
surface generation module which input the profiles formed from prototype of the road detection system; the camera, the laser
projector, the control unit, and the navigation system are to be
901
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008
Barsi, A., Lovas, T., Toth, C.K., 2006 Positioning with mobile
mapping system, in Hungarian, Geodézia és Kartográfia,
Budapest, Vol. LVIII, No. 4, pp. 3-8
902