Research On Detection And Recognition Method For Pavement Crack Based On Feature Space

With the rapid development of highway in recent years, many highways,constructed many years ago, have already begun to be repaired. Somequanlity indicators are used to measure the last work year and maintenanceload of highway, one of which is pavement crack.Now manual detection of pavement crack is the most common method inour country, which is out of date, very slow and dangerous. With thedevelopment of the remote sensing and artificial intelligence technology,non-contacting and vehicle-mounted pavement crack automatic quicklydetection system has been created at home and abroad. This system has manyadvantages such as high-speed, efficient, safe operation and does not affectthe transport, is the most suitable equipment for large-scale highwaypavement crack detection. However, the actual environment of pavementcrack detection is relatively poor, such as uneven illumination, random noiseof the material particles, oil stain, tire marks and the shadow of both sides ofthe trees, which increases the technical difficulty of the pavement crackdetection and identification. So this kind of system has a series of keyproblems such as low accuracy and low efficiency.To solve above key problems and to meet requirements of application,this thesis’s main task is researching in highway pavement crack automaticdetection method, including of learning advanced techniques at home andabroad, designing an appropriate detection method for highway pavement.Below tasks has been done in this thesis.1. Gray image preprocessing methods have been researched, and actualpavement crack images have also been analyzed. In order to solve theproblems which the pavement crack images have, such as the heavy noise and the blurry crack, the fuzzy multi-level median filter algorithm is proposed toenhance the crack image.2. For characteristics of weak information and uncertain thickness ofcrack in the road image and the real-time requirement, a novel pavementcrack detection approach based on weight features is proposed, whichpresents a good detection result.3. A lot of actual pavement crack images have been analyzed andconcluded to classify and model pavement crack preliminary, from which thequalitative and quantitative indicators can be obtained.