Pavement Crack Detection Algorithm Based On Image Analysis

Periodic inspection of Highway Pavement Distress is a necessary part of the road maintenance work of great significance to the maintenance of the public transport safety. With the continuous construction of expressways and high-grade highways, auto-detection system has been extensively studied and applied to pavement damage based on image analysis. Automatic detection algorithm is not perfect for improvement in the efficiency and success rate of the algorithm. With the problems of the previous algorithm, noise reduction and the crack of information enhancement as well as the design of the classifier in these three areas are studied.Proposed an enhancement algorithm based on wavelet transform theory to solve the noise and background issues, combined with the mean filtering algorithm to stably extract the pavement cracks. It is difficult to extract crack disease from picture because of special noise and uneven illumination problem. The wavelet transforms the picture into some levels of high and low frequency information. The road disease could be easily classified by frequency information from background information. Using the template adaptive filter is to smooth the noise information so that crack can separate from background. Using mean filtering algorithm for isolated noise points is a effectively way of noise reduction.Proposed a method of crack regional connectivity to solve the broke road cracks information, which is based on mathematical morphology. Cracks information is relatively weak and excessive smoothing may make some extremely weak part of breaking out. Connection based the mathematical morphology can make the broken part reconnected, and could also strengthen the overall crack which is facilitate to the crack classification.Proposed a crack classification algorithm based on BP neural network to solve the problems of low rate of crack classification. For four different types of crack morphology-horizontal, vertical, diagonal and net cracks, raised three crack characteristics that the horizontal and vertical crack projection, cracks minimum bounding rectangle of crack aspect ratio and crack density characteristics. A number of experiments show that classification algorithm can effectively classify the crack information.