Use of Computer Vision forUse of Computer Vision forCrack DetectionCrack Detection

KrisadaKrisada ChaiyasarnChaiyasarn

supervised by supervised by

Dr Kenichi Soga and Prof. Roberto Dr Kenichi Soga and Prof. Roberto CipollaCipolla

IntroductionIntroduction

Recently graduated with Recently graduated with BA(Hons)/MEngBA(Hons)/MEngin Civil, Structural and Environmental in Civil, Structural and Environmental Engineering from University of Cambridge Engineering from University of Cambridge

MEngMEng project on project on FibreFibre Optics Strain Optics Strain MeasurmentMeasurment (BOTDR) in Reinforced(BOTDR) in ReinforcedConcrete StructureConcrete Structure

Official start the PhD study in Computer Official start the PhD study in Computer Vision in Geotechnical Engineering in Vision in Geotechnical Engineering in October 2007October 2007

OutlineOutline

Current technologies for tunnel inspectionsCurrent technologies for tunnel inspections

Computer vision in Underground MComputer vision in Underground M33

Review of some literaturesReview of some literatures

Possible future planPossible future plan

Current use of imaging techniques for Current use of imaging techniques for infrastructure assessmentinfrastructure assessment

Paris metro: Use Paris metro: Use ofof laserlaser scanner scanner forfor imageimage acquisitionacquisition

SankyoSankyo CoCo. . JapanJapan: : RecordingRecording ofof imagesimages usingusing highhigh--resolutionresolution digital cameras ($18digital cameras ($18--2525-- mm22))

KeisokuKeisoku KensaKensa CoCo. . JapanJapan: Crack : Crack detectiondetection systemsystem: Use : Use ofofhighhigh speedspeed video camera (video camera ($65k$65k//dayday))

Tunnel Inspection System, Takenaka Civil Engineering & Construction Co., Ltd

Focus on Computer Vision (CV) and Machine Learning (ML)

Crack Crack detectiondetection//localizationlocalization

Two steps:

Crack extraction using computer vision techniques (e.g. edge, corner, line, pattern detection)

Crack recognition by machine learning (e.g. support vector machine(SVM) and neural networks)

Above approaches must be together in order to achieve crack detection algorithms

Image processingImage processing

Yu et al. (2006) proposed an inspection system, consisted of a Yu et al. (2006) proposed an inspection system, consisted of a mobile robot with Charged Couple mobile robot with Charged Couple Device(CCDDevice(CCD) for data ) for data acquisition and crack detection system using image acquisition and crack detection system using image processing. processing.

Cracks and non-crack areas are distinguished by their light reflectance values

Cracks are extracted by graph search method with given start and end points

ResultsResults

Erroneous recognition of cracks and nonErroneous recognition of cracks and non--cracks prevailscracks prevailsUsed semiUsed semi--automated system to discard erroneous pointsautomated system to discard erroneous pointsFurther works such as a study of crack characteristics and develFurther works such as a study of crack characteristics and development of opment of fully automated softwarefully automated software

Support Vector MachineSupport Vector Machine

ZhiweiZhiwei et. al.(2002)et. al.(2002) usedused crack crack intensityintensity featurefeature toto detectdetect crack crack edgesedges((DiscriminantDiscriminant AnalysisAnalysis MethodMethod). ). HoweverHowever otherother nonnon--crack crack imageimage may be may be extractedextracted. .

SVM SVM isis usedused toto classifyclassify intointo crack, noncrack, non--crack crack andand intermediateintermediate imageimage. . ExperimentedExperimented by by balancedbalanced andand unbalancedunbalanced subsub--imageimage. .

ResultsResults

Balanced subBalanced sub--image is more effective than image is more effective than unbalanced subunbalanced sub--imageimage

Use image features to enhance image recognitionUse image features to enhance image recognition

OnlyOnly linear linear crackscracks werewere consideredconsidered

SVM SVM needneed furtherfurther developmentdevelopment

Automated detection of cracks in buried concrete Automated detection of cracks in buried concrete pipe imagespipe images (Computer Vision)(Computer Vision)

SinhaSinha et. al.(2005)et. al.(2005) proposes approach for crack detection in proposes approach for crack detection in pipes.pipes.

ResultsResultsResults were good since cracks in pipe images were clearly visible pattern

This paper present excellent system of comparing qualitative values such as visual comparison, correctness, and completeness

Image of mushroom cracks were not as good

How may we go from here?How may we go from here?

ImproveImprove thethe methodsmethods ofof crack crack extractionextraction

ImproveImprove thethe methodsmethods ofof crack crack recognitionrecognition basedbased onon newnewtechniquestechniques suchsuch as as boostingboosting

HighHigh accuracyaccuracy ofof crack crack detectiondetection algorithmsalgorithms toto enableenableevaluationevaluation ofof crack crack sizessizes

TheThe characterizationcharacterization module module wouldwould recogniserecognise crackscracks fromfrom thetheimagesimages andand classifyclassify themthem intointo deteriorationdeterioration categoriescategories

Thank you for your attention