A method of filtering SEM-CC railway track recognition

Abstract

The subject of the solution is a method of SEM-CC filtering for recognizing railway rails located in the space of the track along which the rail vehicle is driven, where the input is the first segmentation mask (SM-1) determining the tracks, rails and background for a certain image frame. Using the baseline model, a second segmentation mask (SM-2) determining the frame of the rails and a third segmentation mask (SM-3) determining the edges of the rails are created. After that, a filtered segmentation mask (SM-4) is created determining the filtered segmentation of the tracks, rails and background so that pixels whose distance from the edges of the tracks and the frame of the rails is less than a preset threshold value are set to the background class. Next, a morphological transformation of the filtered segmentation mask (SM-4) is performed to smooth and simplify the shape of the determined tracks into a morphologically modified segmentation mask (SM-5).Then, all continuous track components within the morphologically modified segmentation mask (SM-5) are determined based on 4-neighborhood, where each component is scored and the component with the highest score is determined as the main track. All remaining pixels of the morphologically modified segmentation mask (SM-5) are set to the background class and the output is the final segmentation mask (SM-6) determining the track and the corresponding rail that is most likely to be the main track.