that the successfully matched covers are confirmed and the
location is calculated from forward intersection with GPS/IMU
data. However, the cover location is taken from the center of
ellipse, it may be a slightly deviation from the actual center of
the cover.
The blue line in fig. 6a shows the missing rate via false
positives (FP) rate use edge and texture information. The
detection rate has a little increase as 88% in 0.05 FP rate for
some round shape objects lack of texture are eliminated. More
ever, the covers are located precisely due to the high accuracy
of GPS and IMU data after matching.
100 I ! ;
= edge information used
— edge and texture information used | ]
90r
80r
missed covers(%)
> an o M
e e e e
T T
A
e
o
T
20r X
1 | i | 1
0 0.05 0.1 0.15 0.2 0.25 0.3
false positive rate
Figure 6. The detection performance based only on edge
information and on edge and texture
3.4 Discussion
This work investigated both detection and location of the
manhole covers using edge and texture information. The test
shows a good performance in covers detection in different
situation under muti-view, occlusions, shadows, sizes.
The simple ellipse fitting method (Chen and Wang, 2006; Liu,
2010) cannot deal with occlusions and false round objects.
More information and process steps should be involved to
decrease the FP.
The single-view and multi-view combined method (Radu and
Luc, 2011) has a high detection rate for more reliable
information such as edges, area, symmetry, been added to
detect and locate covers. However, the detection rate does not
consider the missed covers which cannot be properly segmented
by mean-sift segmentation method.
As mentioned in the above paper (Radu and Luc, 2011), the
RFID tagging base method (Chang et al, 2009) need additional
operation for tagging all the covers and it is not easy to a fast
and convenient browse of all the covers.
Of course, our method also has problems. The biggest problem
is that the following steps all depend on the edge detection
method. In the special circumstances as fig. 3d, it is difficult for
edge detection methods to find proper edges since the intensity
from material of the cover and of the road are very similar. For
the further increasing of detection rate the very weak edge
detection should be solved firstly.
4. CONCLUSIONS
Manhole cover detection is a more challenging problem than
expected that many disadvantageous situations should be
considered. In this paper the vehicle-based cameras combined
with GPS, IMU, LIDAR is used to detect and locate the
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B3, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
manholes in streets. The canny based EDISON edge detection
method and texture based matching method are both used in our
method to assure a high detection rate. The matching result is
also used to accurately calculate the 3D position of covers
combined with GPS/IMU data.
The future work will focus on the weak edge detection and
some other close-range applications using our method, such as
traffic signs detection, traffic lights detection, etc.
Acknowledgements. This work has been partly supported by
Chinese 973 program (2012CB719902).
References
Chang, A.Y., Yu, C.S., Lin, S.C., Chang, Y.Y. and Ho, P.C,
2009. Search, identification and positioning of the underground
manhole with rfid ground tag. Networked Computing and
Advanced Information Management, International Conference
on, pp. 1899-1903.
Chen, Y., Xiong, Z. and Wang, Y.H., 2006. Improved classical
hough transform applied to the manhole cover’s detection and
location. Optical Technique (in Chinese), 32(2006), pp. 504-
508.
Fischler, M.A. and Bolles, R. C., 1981. Random Sample
Consensus: A Paradigm for Model Fitting with Applications to
Image Analysis and Automated Cartography. CACM, 24(6), pp.
381-395.
Liu, J.H., 2011. Research on algorithm of automatically
recognizing and positioning road manhole covers based on
vehicle- mounted sensors. Application Research of Computers
(in Chinese), 28(8), pp. 3138-3140.
Mer, P. and Georgescu, B., 2001. Edge detection with
embedded confidence. IEEE Transactions on Pattern Analysis
and Machine Intelligence, 23(12), pp. 1351-1365.
Radu, T. and Luc, V.G., 2011. Multi-view manhole detection,
recognition, and 3D localisation. [EEE International
Conference on Computer Vision Workshops 2011, pp. 188-195.
Tang, N, 2003. Manhole detection and location for urban
pavement. [EEE transportation system, 2(12), 1552-1555.
LADYBUG, http://www.ptgrey.com/products/spherical.asp. (2
Mar. 2012)
a]
ABS
Digi
after
diffe
sour
usin
imag
laser
mod
ques
the c
base
Ther
algo:
The
infoi
for (
Mod
geos
requ
mod
eval
the i
for
(Rot
Sest
Sour
hum
In te
amo
buil
fron
aeri
data
pres
aeri:
pho
duri
case
com
Syst
the
dor
