A flame detection technique based on fast mean shift and level set method

Abstract

The furnace flame images captured by the CCD are temperature fields inside the furnace. The flame images have some characteristics, such as noisy areas, low contrast and weak edges. The traditional edge detection methods are quick, such as Roberts, Canny etc, but the edges detected are often discontinuous and these methods are sensitive to noise. The level set is a new method to extract geometrical curve which can deal with the geometrical topological changes. It is effective on segmentation of non-rigid objects. But the level set method has some problems: firstly, the computation is very complex; secondly, if the initial curve is given arbitrarily, there would be an increase in iteration times and computational complexity. In view of these features, this paper puts forward a technique which uses fast mean shift and level set methods to segment flame images. Making use of fast mean shift to get the original contours for level set method so that it can decrease iteration times of the level set function. Experimental results indicate that the proposed algorithm can identify the combustion state of a flame more effectively and improve the evolution of the level set function. So it is of extensive practical use.