Abstract:
The collection of pavement surface condition data is usually done by conventional visual and manual approaches, which are very costly, time-consuming, dangerous, labor-intensive, and subjective. These approaches have high degrees of variability, are unable to provide meaningful quantitative information, and almost always lead to inconsistencies in cracking details over space and across evaluations. A novel pavement crack detection approach based on neural network and computer vision, pattern recognition, and image-processing techniques is proposed. The thresholding approach is used to separate crack pixels from the background. The selection of the thresholds is critical to the performance of automated crack detection systems. Statistical values (mean and standard deviation) are used as the features, and they are used to train the neural network for selection of the thresholds. Because of the noise, the resulting images have some isolated spots that can be eliminated by a curve detector. Finally, Hough transformation is used to detect or classify all cracks in parallel. The experimental results have demonstrated that the cracks are correctly and effectively detected by the proposed method, which will be useful for pavement management.
