Generally, inspection of steel bridge coating for decision of repainting time is carried out by mainly observation through inspector's naked eyes, therefore, there are personal equations in inspection results whether inspector has knowledge or experie...
Generally, inspection of steel bridge coating for decision of repainting time is carried out by mainly observation through inspector's naked eyes, therefore, there are personal equations in inspection results whether inspector has knowledge or experience for coating or not. In addition, absolute lack of specialists in inspection of steel bridge coating makes decision of optimal repainting time through objective, quantitative and scientific evaluation more difficult. In order to solve the problem, steel bridge coating diagnosing system using image processing technology was developed in 2003~2004, and this study was carried out to optimize the performance of developed system for field application. From the results of evaluation of casting deterioration degree for 75 bridges in highway, and prediction of coating life through regression analysis, it is predicted to take 13.0 and 13.3 years by exponential and third order polynomial regression respectively to reach deterioration degree point of 70 when coating cannot protect steel any more. According to statistical survey in 2001, average repainting interval of steel bridges in highway was about 10 year. Consequently, it is expected that repainting interval is extended about 3 years, if effective maintenance of bridge coating is carried out by diagnosing system using computerized visual imaging. In order to prepare data base for diagnosing system, atmospheric corrosivity test and survey of climate condition was carried out all over the country. Atmospheric corrosivity with environment was showed as rural < mountains < industry < urban < marine, which is similar to general trend. Corrosivity classes were showed from C2 to C4, and sites corresponded to C3 where corrosivity is medium were most. Through the surbey of climate condition, data base was prepared containing atmospheric temperature, amount of precipitation, UB index, duration of sunshine, etc. For inspection of steel bridge coating using diagnosing system, suitable images are necessary fir image processing. In the study, we standardize photographing method to help inspector take suitable pictures for image processing. In addition, standardization of evaluation area wad carried out by comparison between rust area rate comparing with SSPC standard pictures in field and rust area rate changing reference area to 15×15cm, 20×20cm and 30×30cm by diagnosing system. Optimal standard area was 20×20cm. This research aimed to improve the prototype computer-aided diagnosis system which was developed through previous research in 2003~2004. The prototype system was designed to diagnose automatically painting condition or degree of deterioration of the painting on steel bridge using image processing techniques. The prototype system used hough transform to extract ROI and EM algorithm which is well-known clustering algorithm. However, the diagnosing algorithm of the prototype system was realized to distinguish three regions: rust region, peeled region and normal paint region at one step so it had the drawback which the system couldn't extract the deteriorated region when the color information was distorted by illumination. The improvement was required to overcome the drawback because an investigator can't adjust illumination when he takes pictures of the steel bridge for input of the diagnosing system. We develop new preprocessing and postprocessing of diagnosing algorithm to improve the prototype system in this research. It will allow enough accuracy of the system to use practical work. We add new interface for new ROI setting methodology which is designed for new inspection rule of Korea Highway Corporation. The process which extracts rust and peeled region from image composes of two step. The first step is to extract rust using the Parzen-window which is one of the probabilistic methods. The Parzen window is well-known non-parametric scheme directly uses the samples drawn from an unknown distribution to model its density. The second step is to extract peeled region using EM algorithm and we add post-processing which correct the result of peeled region extraction because the result has region of highlight as peeled region. We added postprocessing algorithm to improve accuracy of peeled region and pain point type rust extraction. Additionally we have done renewal of system interface for more effective display and management of diagnosing results.