Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

72
Detection and Measurement of Pavement Cracking Bagas Prama Ananta

Transcript of Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Page 1: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Detection and Measurement of Pavement Cracking

Bagas Prama Ananta

Page 2: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Overview

Background Aims The Proposed Method Tests and Results Conclusion Future Work

Page 3: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Background Roads are a major asset in most countries To manage these assets, road authorities

need: Accurate, up-to-date information on the

condition of their road network Information on defects is vital to keeping a

well maintain road network

Page 4: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Why do we do a Road Maintenance? Early detection of defects in road surfaces helps:

maintenance to be performed before defects develop into more serious problems, such as potholes and pop-outs.

Thus, detection and measurement of pavement cracking: Provide valuable information on the condition of a road

network Reduce maintenance cost Create a better road network for people to use

Page 5: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Types of Cracks

Transverse Cracking

Page 6: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Types of Cracks…

Longitudinal Cracking

Page 7: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Types of Cracks

Crocodile Cracking

Page 8: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Background…

The 1st maintenance process is the detection of defects

Once detected, defects can be analysed and a decision can be made as to what action needs to be taken

Page 9: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Present Method

Visual inspection Two operators travel at 20 km/h One as the driver, another to record the defect Time consuming, costly and can be

dangerous

Page 10: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Present Method… An improved method

A video based system Able to record the pavement up to 100 km/h The recorded video is then inspected off-line

at speed of 20 km/h

Page 11: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Present Method…

Page 12: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Project Aims

Proposing a method of semi-automated detection of cracking defects in the road pavement from video footage.

Advantages of a semi-automated system: Faster More reliable More accurate

Page 13: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Challenges Low resolution of the captured image

768x576 pixels or 0.44 megapixels Lossy compression is used

To make storage of the data practical Highly variable lightning conditions Potential false identification of cracks

Shadows, rail and tram tracks, other road objects

Page 14: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Challenges

Sample set provided by PureData, however the images were not suitable for testing. Resolutions are too low Most images are not sharp (i.e. a lot of blurry

images) which result in noises 1200x900 (~1mp) images are used to test

the method

Page 15: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Commercial Implementation

Several companies offer solutions for monitoring road surface condition

Such solution are the CSIRO and Roadware crack detection systems

Due to the commercial nature, information on their operation is limited

Page 16: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

CSIRO’s Road Crack Detection Vehicle

• Comprised of mostly custom designed and manufactured hardware• The system is very expensive and requires specialised maintenance

Page 17: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

CSIRO’s Road Crack Detection Performs all data analysis in the field No image data is kept The only output is the road quality report Leads to uncertainty with the accuracy of the

results Further manual inspection is needed to

guarantee the results of the systems

Page 18: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Roadware’s Wisecrax

Performs all data analysis off-line

Dual video cameras record 1.5 m by 4 m sections of pavement

High intensity strobe lights produce consistent illumination of pavement images

Page 19: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Solution to Similar Problems Crack Detection by the use of a laser based system Work on this problem was commenced by a

previous honours student (Timothy Evans). A modified watershed algorithm was proposed Difficulty in testing his algorithm This project uses part of Tim’s method for detecting cracks

Sun et. al [2] proposed a new segmentation algorithm for detecting tiny objects Edge detection, line growing and line cutting

Crack detection based on the “grid-cell” analsyis by Xu and Huang

Page 20: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

The Proposed Method

To use image processing techniques to segment the cracking information. Seed Selection Line growing Noise removal

Page 21: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Pipeline of Solution

Page 22: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Initial Detection or Seeding

Horizontal and Vertical Scan Contrast Comparison Combine seed

Page 23: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Profiles of Cracks, Lane Marks and Shadows The challenge in crack detection is to differentiate

between cracks and noises, where noises are: Stone texture Leaves, branches, etc Lane Markings Shadows

Analysing the different between the profiles between cracks and noise (lane marks and shadows) is useful for segmenting the crack from images.

Page 24: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

A Lane Mark profile

Page 25: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

A Shadow Profile

Page 26: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

A Crack Profile

Page 27: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Cracks on Shadows

Cracks

Cracks on shadows

Page 28: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Seed Selection - Horizontal and Vertical Scan

Page 29: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Original Image

Page 30: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Horizontal and Vertical Pass

Page 31: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Seed Selection – Contrast Comparison

A represents the current pixel. B and C are the candidate pixel for growing.

Calculate all the 4 directions: R=max(R(a), R(b), R(c), and R(d)).

If R > T, then the seed is validated else seed is discarded

Page 32: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Original Image

Page 33: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Contrast Selection

Page 34: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Seed Selection - Combination The proposed method of seed selection The combination of Horizontal & Vertical

and Contrast comparison More accurate

Page 35: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Original Image

Page 36: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Horizontal and Vertical Pass & contrast Selection

Page 37: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Line Growing – Watershed transformation current pixel

• Start from the current pixel

• Mark pixels that are similar to the current pixel as a potential crack seed

Page 38: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Original Image

Page 39: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Watershed Transformation / Line Growing Algorithm

Page 40: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Noise Removal

Flooded points must not be too close with each other to the extent the area is overcrowded

A crack will generally have a certain width A crack will generally not be an isolated

pixel

Page 41: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Original Image

Page 42: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Noise Removal – Over Crowded

Page 43: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Original Image

Page 44: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Noise Removal – Isolated Pixels and Crack Width

Page 45: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Original Image

Page 46: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Horizontal and Vertical Pass

Page 47: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Contrast Selection

Page 48: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Horizontal and Vertical Pass & contrast Selection

Page 49: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Watershed Transformation / Line Growing Algorithm

Page 50: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Noise Removal – Over flooding

Page 51: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Noise Removal – Isolated Pixels and Crack Width

Page 52: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Currently using a global threshold to determine the seeds

During seed selection varying lightning condition make the selection of a global threshold difficult

Solution: a localised threshold method is proposed

Inconsistent Lightning Condition

Page 53: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Original Image

Page 54: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Result using a global threshold

Page 55: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Result using a localised threshold

Page 56: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Original Image

Page 57: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Result using a global threshold

Page 58: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Result using a localised threshold

Page 59: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Original Image A Result of Image A

Original Image B Result of Image B

Page 60: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Original Image - Horizontal Crack

Page 61: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Result of the original image - Horizontal Crack

Page 62: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Original Image

Page 63: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Result of the Original Image

Page 64: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Original Image

Page 65: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Result of the Original Image

Page 66: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Test and Results

The algorithm is tested over 123 images Images of pavement containing cracks (67)

95.5% of successful detection 4.5 % of false detection (due to inconsistent lightning)

Images of pavement containing no cracks (56) 64% of successful non detection 36% of false identification of crack (due to road edges,

shadows on leaves and stick)

Page 67: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

An pavement image containing no cracks

Page 68: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Result of the Original Image

Page 69: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

An pavement image containing no cracks

Page 70: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Result of the Original Image

Page 71: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Conclusion and Future Work Project Result:

A semi-automated crack detection system Works with 1megapixel images Achieve 81.3% of success Achieve 18.7% of failure

More work on the seed selection process using a localised threshold Test other techniques for noise removal:

Supervised Learning – recognition of crack and noise patterns

Page 72: Detection and Measurement of Pavement Cracking Bagas Prama Ananta.

Any Question?