A novel trouble of moving EMU detection algorithm based on contextual semantic information

It is essential to detect the state of EMU´s component while running, since any small and subtle failure may cause major accidents in high-speed running. The traditional detection approach adopts the image matching technology, which suffers from the problem when the two images dislocation. The drawback stems from the image matching approach based on the visual features only easily affected by the image quality and transformation. To overcome this defect, this paper introduces the contextual semantic information into image matching technology to detect the trouble of moving EMU. There are two areas of novelty: first, the useful contextual semantic constrained information is obtained by Conditional Random Fields model instead of by manually labeled. And then we combine the feature appearance similarity and contextual semantic information together in order to match more accurate. The experimental result has shown that the proposed algorithm can detect the trouble of moving EMU effectively, even in the circumstance with low image quality, uncontrollable light and bright.