Research On The Key Technologies Of Pavement Crack Detection Based On Image Analysis

Crack is the most common diseases in the pavement, while it is an early form of the vast majority of disease. To detect cracks and repair road surface timely can minimize disease and reduce losses. In addition, cracks are often more subtle and the crack detection is the difficult problem in automatic disease detection, so this study aimed to crack disease detection. In this dissertation, the fractional differential, multi-scale transform theory, median filtering, image morphology and morphological component analysis applied to the road crack image enhancement, denoising, edge detection and shadow separation, the goal is to further improve crack detection and shadow separation and promote the development of automatic crack detection.First, pavement crack detection based on fractional differential and image morphology is proposed. After analyzing the amplitude-frequency characteristics of the signal, it is found that fractional differential can enhance the high and intermediate frequency part of the signal and retain non-linear low frequency part. And integer-order differential can enhance the high and intermediate frequency part but at the same time weaken the low frequency part signal. According to the classical definition of fractional derivation G-L, derive the definition of fractional differential, build a fractional differential mask, and enhance road image based on the fractional differential mask operations so that weak cracks signals in the smooth region are effectively strengthened. Then in order to filter isolated noise and detect crack, enhanced image is operated by image morphology operators and a group of median filters. Experimental results proved that the algorithm is more robust and effective to detect road cracks especially for weak contrast cracks and thin cracks than traditional algorithms. Meanwhile, the fractional differential operators make up the defect of the traditional image enhancement operators that can not change the parameters to get the enhanced effect of continuous change, and for the weak signal processing is not satisfied with the defect. The algorithm is more flexible and targeted. And higher operating efficiency of the method can be widely used in the road image disease testing.The characteristics of multiscale geometric analysis are multi-scale, time-frequency localization and multi-directional. Contourlet transform, a kind of multi-scale, multi-dimensional expression, can accurately grasp the image geometry information, effectively capture the natural contours of the image, but there is spectral aliasing, thus weakening the direction selectivity. NSCT canceled the sub-sampling part in contourlet transform, multi-scale decomposition by the nonsubsampling tower filter, and then decomposition sub-band image under the nonsunsampling directional filter getting the sub-band images in different scale and in different directions. NSCT transform can accurately grasp the linear features, and can well inhibit the pseudo-Gibbs artifacts in the crack edge blur. First, the coefficients at different scales and in different directions are obtained by image decomposition using the NSCT, then with these coefficients thresholds are adaptively set and the generalized nonlinear gain function is used to enhance the features with low contrast while protecting the strong contrast features from over enhancing in the NSCT domain. After the enhancement, reconstruction of these coefficients is performed. Last, morphological operators and median filters are used to detect cracks and remove isolated noise. This algorithm is tested by actual highway pavement images. The results demonstrate that the algorithm is more effective to detect road cracks especially for weak contrast cracks and thin cracks than other algorithms.Shadow is the area that light is completely or partially blocked by the camera. The gray values in the shaded area are smaller than that in the surrounding area. The gray value of cracks is lower too. So the shadow will affect the accuracy of road crack identification, and increase the rate of error detection。For the shadow problem, shadow separation algorithm based on morphological component analysis is presented. The algorithm is related to the sparse representation theory and morphological component analysis. The key assumption of morphological component analysis is that the image geometry and texture components are sparse in a particular library or over-complete dictionary, and they are irrelevant. First, according to the geometric characteristics of the road image and shadow find the appropriate dictionary to sparse represent every part. Then, transform the corresponding image signal according to the dictionary to get the sparse coefficients of the various parts information, and the coefficient should shrink with a soft threshold to the new one. Finally, the inverse transform of the coeffiencient. This process is iterative to separate the shadow information from the road image. The experimental results confirmed that the shadow separation method proposed in this dissertation is effective.