DSP Algorithms for the Automatic Inspection of Fixing Devices of Railroad Lines

摘要

The inspection of railway systems has always been an expensive task that, up to now, has been performed by persons without any substantial assistance. In the future, it will be extensively automated through the analysis of video recordings by means of digital signal processing and pattern recognition methods. One of the inspection tasks is to verify the conditions of rail fastenings. In the present paper, principles of methods are introduced which make the automatic recognition of railway clamps and their components possible. Also, various aspects of image preprocessing and normalization, matched-filtering, feature extraction, and classification by neural networks are discussed with respect to the above mentioned applied problem.The complexity of the problem described in this paper is conflicting with very small frame rates of the on-line analysis unit and with requirements for a robust, reliable, and commercially acceptable signal processing system. The proposed solutions are designed to meet these requirements in an effective way.