Computer vision-based Road Crack Detection Using an Improved I-UNet Convolutional Networks

Cracks are one of the common and important diseases on the pavement surface. The traditional crack detection method mainly relies on manual operation, which is time-consuming and laborious in practical treatment. Therefore, an improved road cracks algorithm based on the I-UNet is proposed in this paper. In the method, the dilated convolution is used to expand the receptive field of the convolution. The method likes "Inception" is used to extract different scales of image features and conduct multi scale feature fusion, the Elu activation function is used. After training, I-UNet can efficiently segment complete cracks in complex environments. Experiments show that the proposed I-UNet based road crack detection method is more robust and more accurate than the U-Net based method.

[1]  Trevor Darrell,et al.  Fully Convolutional Networks for Semantic Segmentation , 2017, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[2]  Yichang James Tsai,et al.  Implementation of automatic crack evaluation using Crack Fundamental Element , 2014, 2014 IEEE International Conference on Image Processing (ICIP).

[3]  Oral Büyüköztürk,et al.  Deep Learning‐Based Crack Damage Detection Using Convolutional Neural Networks , 2017, Comput. Aided Civ. Infrastructure Eng..

[4]  Manuel Avila,et al.  2D image based road pavement crack detection by calculating minimal paths and dynamic programming , 2014, 2014 IEEE International Conference on Image Processing (ICIP).

[5]  Qingquan Li,et al.  CrackTree: Automatic crack detection from pavement images , 2012, Pattern Recognit. Lett..

[6]  Wei Wang,et al.  Computer vision-based concrete crack detection using U-net fully convolutional networks , 2019, Automation in Construction.

[7]  Thomas Brox,et al.  U-Net: Convolutional Networks for Biomedical Image Segmentation , 2015, MICCAI.

[8]  Jérôme Idier,et al.  A new minimal path selection algorithm for automatic crack detection on pavement images , 2014, 2014 IEEE International Conference on Image Processing (ICIP).