Modified random walker segmentation method of welding arc thermograms for welding process diagnostics

Machine vision systems are being used more and more often in many branches of industrial engineering and observation of welding is beginning to have an important role. Visible light and infrared cameras are applied in order to acquire images of already made joints to control their quality, but the area of the welding arc and the welding pool of molten metal are also observed. The infrared cameras are regarded as the most useful. In the paper an approach to the segmentation of welding arc thermograms with the use of a random walker based technique is presented. The random walker algorithm is an algorithm for image segmentation that needs seeds for all segments. In the proposed approach a new and original edge weighting function was introduced. The new weighting function describes the specificity of the thermographic images better, similarly to the case of other commonly used functions. In addition, the automatic seed selection procedure is described. The results obtained show that the modified random walker algorithm can be successfully included in the image processing path that can be a part of a diagnostic system used on a manufacturing line.