Bogazici University Pattern Analysis and Machine Vision Laboratory |
|
|
|
BRITISH COUNCIL ACADEMIC LINKS PROGRAM Bogazici University-Istanbul & City University-London STEEL STRUCTURES: AUTOMATIC CLASSIFICATION OF RUST AND SURFACE CLEANLINESS - REPORT No.1/May 1996
Steel bridges, such as those spanning the Bospherous River, are valuable economic and social assets. To ensure long useful lives for such structures, regular inspection and maintenance are essential. An important part of this is to know when the protective coatings must be renewed. Failure to do so can result in a bridge becoming unserviceable and needing multi-million dollar repairs. New technology for assessing the condition of the protective coatings, and the possible extent of underlying corrosion, is the subject of a joint research programme recently started between Bogazici University and City University, London. The objective is to work out methods for large scale, automatic classification and thus overcome the current dependency on human judgement and limited sampling in condition assessment and preparation for re-coating. The degree of rust and the condition of the steel surface after the application of the cleaning process are currently judged by comparing the surface to a relevant visual standard. Examples of these are shown below as two grades of rust (C & D) and two intensities (2 & 2 1/2) of Ultra-High Pressure Water Blasting (UHPW), an environmentally friendly cleaning method for steel structures. Unfortunately, even small variations in the direction of the sun can alter the appearance of the surface and affect human judgement. For this, and other reasons, the use of image processing is being investigated. Texture analysis plays an important role in automatic visual inspection of surfaces. The computer vision team at Boğaziçi University have used several techniques (features from concurrence matrices, wavelet transforms, a variation of the Karhunen-Loève Transforms, AR filters, Markov Random Fields) for inspection of various textured surfaces. Based on some preliminary work, the most promising technique for the examination of steel surfaces seems to be a model based texture analysis method where the texture model is a Gauss Markov Random Field model. Model based texture analysis methods try to capture the process that generated the texture. They try to model the texture by determining the parameters of a predefined model. The brightness level at a point in an image is highly dependent on the brightness levels of the neighbouring points unless the image is simply random noise. Markov random fields use a precise model of this dependence. They are able to capture the local (spatial) contextual information in an image. These models assume that the intensity at each pixel in the image depends on the intensities of only the neighbouring pixels. The intensity of illumination may not be critical for this type of modelling. Using this approach, the preliminary results indicate that it is possible to differentiate the following:
| |||||||||||||||||||||||
Please send mail to dogozgur@boun.edu.tr with questions or comments .
|
