Research On Recognition Algorithm Of Pavement Cracks

Detection of pavement cracks has been playing a key role in the pavement maintenance. Traditional detection methods which based on human vision can not meet the requirements of the pavement development. The traditional human-vision detection has a lot of drawbacks, such as high-cost, low-efficiency, high-dangerousness, affecting transportation and inaccuracy. Therefore, it is needed to find a new method to improve detection efficiency. At present, the new detection method which is based on digital image processing technique has become a hotspot in the recognition of pavement cracks all over the world. In contrast with the traditional method, it uses computers to process the pavement images and collect the information of pavement cracks. However, due to the complexity and diversity of pavements, most of the detection algorithms can't fit all of the pavements.In order to address the above task, this article proposes a novel algorithm which can adapt to different pavement surface and various light condition of pavements. Firstly, we use image pre-processing operations to enhance images. The Following BP neural network operation can identify the regions which contain cracks. Ultimately cracks information can be extracted from the images.In the image pre-processing operations, the median filtering and Fourier transforms are used to eliminate the image noise. Then the partially overlapping sub-block histogram equalization method can reduce the effects of uneven illumination. Finally the high-frequency-emphasis filter effectively enhances images and highlights the details of the image characteristics.Based on the above pre-processing of images, BP neural network is used to recognize the cracks in these images. First of all, the original images are divided into some smaller square sub-blocks. Combined with the feature of sub-block which contains cracks, characteristic parameters are extracted. Then according to these characteristic parameters, BP neural network determines if theses cracks existed in a sub-block. At last, Hough transform is used to remove the cement joints noise.Ultimately, taking use of the results of previous processing, we can extract the useful information of pavement cracks. By supposing a set of conditions, the connected domains of cracks can be extracted. The following open-close operation is used to remove burrs, hollows and gaps. Then skeleton algorithm can get cracks skeleton information and calculate the width and length of the cracks.