Research On Automatic Pavement Distress Detection Based On Image Analysis

Pavement distress detection plays an important part in maintaining pavements. Traditional manual methods can't meet the need of pavement's development now, because they have the following shortcomings such as time-consuming, non-precision, dangerous, costly and also affecting transportation. Therefore, it is necessary to propose a kind of fast method to improve the detection's efficiency. Approaches based on images have become the hotspot in this fiend over the world, which use high rapid and precise cameras to screen pavement images, with fast processing of computers, to get distress information. However, because of road surface images' complexity and diversity, and also distress information's weak-signal, it is not easy to get efficient detection algorithms. There exist many problems including imaging quality, lamp, shadow, stability and etc.In this paper, a new image gray correct method is proposed in order to solve gray un-uniform. The effect of illumination's un-uniform is evident on the whole image, but a local part in the image can be thought uniform. At least, the image's background or object can be uniform or approximately uniform in some small areas. Therefore, making use of segment method, extracting the point sets that reflect the changing trend of image gray; then, getting the background image with interposing means; finally, running operations on original and background images, the correct image can be had. Experiment result shows this pre-processing way removes the gray un-uniform in the image effectively, and ensures the following distress detection going on well.Aiming at pavement crack distress's detection, this paper advances two different methods. The first one makes use of the image enhancing approach investigated in this paper to do the whole gray correct; then, uses distress's statistic characteristic in the image to get one threshold to segment image, extracts every object block's linear features which are used to recognize linear object block; at last, distinguishes the true or false blocks. The second method is based on multi-fit mode. Extracting possible crack points of image blocks; then drafting the lines that accord with the crack skeleton's direction. Fusing the results of several overlap windows can get the nearly intact crack object. Lots of sample experiments under many kinds of conditions have proved the efficiency of the method. Its precision can arrive at 2mm.The collection of pavement distress images is a sequence collecting process. There arealways overlap parts in the consecutive images, which must bring forward some distress' overlap. One advantage of mosaiced image is its intuitionistic effect for human vision. Putting the detected distress into a larger result image can make it easy to get the distress' distribution in the whole pavement. Aiming at mosaice the distress images and results, here a mosaicing method based on distress results is proposed. But there are not enough features of pavement surface image, it is difficult to attain accurate matching. Pavement distress has relatively more information and is easy to match fast. For the overlap parts of consecutive images, finding the matching areas according to distress information, so can mosaice all distress images and their result images.At last, an automatic detection system on pavement distress is presented, including pavement images collection system and off-line process system. This system is the first developed one in our country. At present, it has been used and completed several highways' detection tasks. The final experiments and practical using result show this system has reached the world level and meet the need of pavement maintenance in the country.