Pixel level pavement crack detection using deep convolutional neural network with residual blocks
Road condition monitoring, such as surface defects and pavement cracks detection, is an important task in road management. Automated road surface defect detection is also a challenging problem in computer vision and machine learning research due to the large variety of pavement crack structures, variable lighting conditions, interfering objects on the road surface such as trashes, fallen tree leaves and branches. In this work, we develop a deep learning-based method for automated road surface defect and pavement crack detection. We design a deep convolutional neural network based on using residual blocks to predict the heatmaps which indicate the location and intensity of defects and cracks. To reduce false detection rates, we couple this heatmap prediction network with a binary classification network which is able to determine if the input image patch is normal or has defects. We test our method on the CFD benchmark dataset. Experiment results show that the proposed network is very effective for pavement crack detection and has more advanced performance than other methods.
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 License.