Abstract:
In this paper different image processing and computer vision techniques are applied in order to identify and segment pavement cracks presented in digital images from a public repository. Four distinct pipelines are presented and compared, three of them started with image pre processing for contrast increase and noise reduction. The last one was only composed of a U-Net neural network, frequently used in medical image segmentation. The neural network was trained and presented good results despite of illumination differences and other image noise sources. A drawback observed was the need of a great number of images in a well balanced and not biased data base. Finally, parameter extraction is discused for future crack classification.
Reference:
MARTINS, Gabriel Borelli; KLEINE, Felipe Augusto Frazão; BERNARDI, Ely; VIEIRA, Rubens; MACHADO Aline Ribeiro; RODRIGUES, Renato Curto; SOARES, Elaine Maria. Método de segmentação de imagens digitais para identificação de trincas em pavimentos viários: um estudo comparativo. In: ENCONTRO NACIONAL DE CONSERVAÇÃO RODOVIÁRIA, 23., REUNIÃO ANUAL DE PAVIMENTAÇÃO, 46., 2021., virtual. Anais… 13 p.
Document with restricted access. Log in to BiblioInfo Biblioteca-DAIT/IPT to access the PDF text:
https://escriba.ipt.br/pdf_restrito/177564.pdf
In this paper different image processing and computer vision techniques are applied in order to identify and segment pavement cracks presented in digital images from a public repository. Four distinct pipelines are presented and compared, three of them started with image pre processing for contrast increase and noise reduction. The last one was only composed of a U-Net neural network, frequently used in medical image segmentation. The neural network was trained and presented good results despite of illumination differences and other image noise sources. A drawback observed was the need of a great number of images in a well balanced and not biased data base. Finally, parameter extraction is discused for future crack classification.
Reference:
MARTINS, Gabriel Borelli; KLEINE, Felipe Augusto Frazão; BERNARDI, Ely; VIEIRA, Rubens; MACHADO Aline Ribeiro; RODRIGUES, Renato Curto; SOARES, Elaine Maria. Método de segmentação de imagens digitais para identificação de trincas em pavimentos viários: um estudo comparativo. In: ENCONTRO NACIONAL DE CONSERVAÇÃO RODOVIÁRIA, 23., REUNIÃO ANUAL DE PAVIMENTAÇÃO, 46., 2021., virtual. Anais… 13 p.
Document with restricted access. Log in to BiblioInfo Biblioteca-DAIT/IPT to access the PDF text:
https://escriba.ipt.br/pdf_restrito/177564.pdf
