Automatic segmentation of pavement condition data using wavelet transform

We present a new algorithm based on wavelet transforms for automated segmentation of pavement (road surface) condition data. We developed a denoising scheme to remove random noise caused by the collection device and random extreme distress in the pavement while essentially preserving the important information followed by a singularity detection-based segmentation algorithm. During the segmentation stage, singularities of the smoothed waveform are detected, and they are marked either as isolated singularities or border points. Isolated singularities are suppressed and the remaining singularities are used as border information to segment the pavement-condition data into regions that exhibit similar characteristics. The proposed approach follows the envelope of the original pavement data and the resulting segments semantically and objectively (in the mean square error sense) represent the original data better than the previous methods.