Study On Multi-level Denoising And Crack Detection With Concrete Surface Image

The rapid development of road bridge engineering promotes the operation and maintance technology to make a great progress.Cracks are common distress in maintance that may potentially threaten the road bridge structure safety.Traditionally,cracks have been identified by human inspectors who use the simple equipment through visual surveys.But these manual inspections are costly and time-consuming.Since image processing based approaches provide an efficient and economical way for crack detection,automated crack detection has gradually replaced manual inspections and become development trend.Concrete surface images crack detection methods were studied in this thesis.With the advantange of percolation model detecting unclear and tiny cracks,dark pixels extracted by dividing overlapping windows was used to accelarate percolate.Multi-level denoising process was used to improve the accuracy of crack detection.Meanwhile,a robust crack detection method which fast extract crack edge using structured forests was proposed.The main research work is as follow:1.For untreated concrete surface image,there are many kinds of noise superimposed on target crack information,which results in the low accuracy rate of crack detedtion.The characteristics of noise were studied to achieve multi-level denoising.To make the crack information is more conspicuous constrast with the background.2.Aiming at the problem of low efficiency of crack detection using percolation model,a divided overlapping window to extract seed dark pixels method was proposed.The velocity of percolation process was accelarated by only percolating the seed dark pixels.Combined with the multi-level denosing method,a fast percolation detection algorithm based on image multi-level denoising was proposed.The accuracy of percolation algorithm was improved.3.A structured forest learning framework was used to crack detection.Local structured information was used to learning decision trees and extracted crack edge information effectively.The result was robust for detecting crack edges on different datasets.Then,edge information is used to percolating which can accelarate the percolation process.A percolation detection algorithm with structure forest edge detection was proposed.