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

With the expanding of highway network in our country, the scale of the highway construction is becoming more and more leveling off, it is an important task to detect, maintain and mamage the pavement in the highway construction field. 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. Therefore, the pavement crack detection has important practical significance.For complex textures, object types, object signal and illumination in images of pavement, it is harder to identify them than other images. The existing detecting algorithms are developed on the pavement image with good quality and clear cracks, which is not suitable to complex environment, it is difficult to meet the needs of engineering applications.Focusing on the problems above mentioned, this study had researched on noise removing, target segmenting, cracks edge extracting and type recognizing etc, and then proposed an intelligent pavement crack detection method based on image processing. It established the compressive perception denoising model on gray image of the pavement crack. For the asymmetrical illumination of images, the adaptive threshold model was builded for the crack segmentation, the edge segmentation threshold was calculated according to the Weber law and visual model and, finally identifies the types of the cracks in images by the radial basis probabilistic neural network.This research has fruits and innovation as follows:1. According to the imbalance between signals and noises, a novel compressed sensing filtering algorithm was proposed in Nonsubsampled Contourlet Transform(NSCT) domain. The NSCT was adopted to conduct the multi-scale and multi-direction decomposition on the noisy pavement distress image. The decomposition results can give the low-frequency coefficients and the high-frequency coefficients, the compressive perception denoising model is built, the high-frequency coefficients were observed by the pseudo-random Fourier matrix. By building the optimization objective function in reconstruction process and the parameter setting, the different optimization objective function signal was applied, the original image was reconstructed to achieve the purpose of realization of filtering. Simulation results show that the proposed algorithm can be very good to overcome the influence of uneven illumination target detection of cracks, compared with the Curvelet and Wavelet threshold denoising, the error detection ratio of the cracks decreased 15% and 7.8% respectively under the same simulation conditions.2. According to imaging illumination uneven problem, an adaptive threshold segmentation algorithm based on the improved FCM was put forward. The fuzzy histogram was used in the improved FCM for pavement gray image, the number of fuzzy histogram peaks were used to determine the clustering number, the maximum and minimum values of fuzzy histogram were adopted to determine the clustering center. The estimated global segmentation threshold was obtained by the improved FCM algorithm for the whole image, OSTU algorithm was used to estimate the local segmentation threshold for the sub image, the actual local segmentation threshold was determined jointly by the local estimates threshold segmentation and the relationship between the global estimate threshold segmentation and sub image grayscale characteristics. Simulation results show that, the recall ratio of the adaptive threshold segmentation algorithm was proposed in this paper increased by 0.195, the precision ration improved by 0.0566 compared to the FCM algorithm.3. The pavement crack edge detection threshold algorithm based on the visual model was proposed. The opening and closing operations based on multi-scale contour structuring elements were used to remove noise in this paper, and grade image was obtained by the smallest size contour structuring. An image can be divided in to low dark area, middle area and high bright area through Weber’s law according to the bionics principle and background brightness. The edge thresholds for the different areas were calculated according to the different equations respectively. The proposed crack edge detector was tested on popular images having different image properties and also compared with popular edge detectors from the literature. Simulation results show that, the proposed algorithm was less affected by noise and it was able to overcome the influence of uneven illumination, can detect the edge cracks in relatively complete information, the proposed method had the better quantitative indicators compared with the edge extraction based on the multi-scale wavelet modulus maxium and pulse coupled neural network under the same experimental conditions.4. A classifier based on the Radial Basis Probabilistic Neural Network(RBPNN) was designed to identify types of cracks. The radial basis function was used as the base functions in first hidden layer, the k means clustering method was used to calculate the implicit vector center, the connection weights between the second hidden layer and the output layer was determined by recursive least squares method. The geometrical characteristics of the cracks were chosen as network input, 900 known fracture types of pavement image were used as the training sample for the radial basis probabilistic neural network, then 700 unknown types of pavement image were tested by the trained RBPNN. The test results show that the radial basis probabilistic neural networks can achieved classifying effects satisfactorily, the recognition accuracy increased 1.85%, 4.35% and 7.78% respectively compared with SVM, C4.5 decision tree classifier, the bayesian classifier.