We incorporate, evaluate, and assess the feasibility of
using filter banks in automated pavement distress systems from a
system level. We integrate a novel filter-bank-based distress segmentation
method, which, unlike previously researched methods, does not
depend on highpass data. In addition, we incorporate the standard
Said Pearlman set partitioning in hierarchical trees compression
coder into the automated pavement distress system, which is a
first in this area of research. A third contribution of the research is
a statistical detection algorithm that assists in overall system performance.
Preliminary testing using images provided by the Georgia
Department of Transportation demonstrate the promise of the proposed
method.
1. The Center for Signal & Image Processing Georgia Institute of Technology
Enhanced adaptive filter-
bank-based automated
pavement
crack detection and
segmentation system
By
Clyde A. Lettsome, Yi-Chang Tsai, and
Vivek Kaul
2. Outline 2
The Center for Signal & Image Processing
• Background
• Design Challenges
• Proposed System
• Results
• Conclusion
3. 3
Background The Center for Signal & Image Processing
• Most of the state departments of transportation
(DOT) use either visual or manual distress
inspection systems, which are costly, dangerous,
time-consuming, labor-intensive, and subjective.
• Desire – Develop effective and cheap automated
pavement distress system collects pavement
images or video and detects distress without
human intervention.
4. Background 4
The Center for Signal & Image Processing
• Zhou1
proposed a
popular
automated
distress
detection and
segmentation
structure with
two main
sections.
5. 5
Background The Center for Signal & Image Processing
• Popular Filter-bank-based systems.
• Zhou1 proposed distress detection method that
compared the nonzero values in the highpass subbands
to predetermined thresholds.
• Li2 proposed a distress segmentation method that
combined threshold selection method of Mallat and
Zhong3 with Gaussian filtering to remove noise and
detect edges in images.
• Advantage filter bank methods allow both spatial
and frequency domain analysis.
6. 6
Background The Center for Signal & Image Processing
• Disadvantages to both proposals.
1. Filter-bank decomposition, distress detection done on
highpass data. Overlap and add due to row and
column filtering causes construction and
destruction of highpass data.
2. If standard compression coders (S+P SPIHT coder or
JPEG 2000), segmentation would be performed on
degraded high-low, low-high, and high-high
subbands.
7. 7
Design Challenges The Center for Signal & Image Processing
Pavement Distress
Row 140 of Pavement Distress
Image
image
9. 9
Proposed System: Preprocessing The Center for Signal & Image Processing
•Values larger than the
mean minus one
standard deviation are
normalized to the
mean of the image.
•Other values remain
the same.
An image preprocessed to remove
surface texture.
10. 10
Proposed System: Time-Varying Filtering The Center for Signal & Image Processing
Complimentary filters
• G00(z) low-delay lowpass filter
•G01(z) linear-phase lowpass filter
• G02(z) high-delay lowpass filter
Proposed System: Time-Varying Filtering
11. 11
Proposed System: Time-Varying Filtering The Center for Signal & Image Processing
Why
these
filters?
(a) Low-delay lowpass filter step response
(b) High-delay lowpass filter step response.
12. 12
Proposed System: Time-Varying Filtering The Center for Signal & Image Processing
An internal block diagram of the time-varying filtering block.
13. 13
Proposed System: Segmentation The Center for Signal & Image Processing
A window
function of Li ×
Li, where Li is
the length of the
linear phase
filter used in the
development of
the mask.
An edge detection mask developed from row filtering.
14. 14
Proposed System: Clustering and HVS The Center for Signal & Image Processing
• Since current ground truths are determined
empirically it is important to consider the human
visual system (HVS).
• Relationship between intensity and brightness is
not linear.
• Ernst Weber4 found that a perceived change in
intensity occurs when
15. 15
Results GDOT image #1D579384 The Center for Signal & Image Processing
(a) Ground Truth (b) Modified filter bank (c) Li/ Mallat and Zhong
16. 16
Results GDOT image #1D579384 The Center for Signal & Image Processing
(a) Ground Truth (b) Modified filter bank (c) Li/ Mallat and Zhong
17. 17
Results S + P SPIHT Compressed Images The Center for Signal & Image Processing
(a) GDOT image #1D579384 (a) GDOT image #1D579384
18. 18
Conclusion The Center for Signal & Image Processing
We focused on incorporating, evaluating, and
assessing the feasibility of using wavelet/filter banks
from a system level.
The advantage of the proposed method is
that, despite the compression rate, it can be used on
raw or compressed images.
The proposed system exhibited significant
improvement versus existing filter-bank-based
pavement distress segmentation methods.
19. 19
Bibliography The Center for Signal & Image Processing
1. J. Zhou, P. S. Huang, and F.-P. Chiang, “Wavelet-based pavement distress detection and
evaluation,” Opt. Eng. 45(2), 027007 (2006).
2. J. Li, “A Wavelet Approach to Edge Detection,” Master Thesis, Mathematics Sam
Houston State University, Huntsville, Texas (2003).
3. S. Mallat and S. Zhong, “Characterization of signals from multiscale edges,” IEEE
Trans. Pattern Anal. Mach. Intell. 14(7), 710–732 (1992).
4. M. J. T. Smith and A. Docef, A Study Guide for Digital Image Processing, Scientific
Publishers Inc., Riverdale, GA (1999).