Obstacle detection in railway tracks: vehicle autonomous system application
Swagata B. Sarkar, Leo John Baptist, K. Valarmathi, Sheshang Degadwala · International Journal of Vehicle Autonomous Systems · 2025
For the purpose of preventing accidents on railroads that might result in casualties and perhaps cause the train to be damaged or derailed, reliable obstacle detection could be of great assistance. The present system has drawbacks such as the generic object detectors do not have sufficient classes to account for all of the conceivable circumstances, and it is difficult to acquire data sets that include items on trains. This paper proposes the use of a shallow network for the purpose of learning railway segmentation from regular railway photos. The small receptive field of the network avoids overconfident predictions and enables the network to concentrate on the locally extremely distinct and recurring patterns of the train environment. Our system novelty gives the network the ability to maximise its effectiveness. The approach that we use is evaluated using a specialised data set that contains railways images that have been intentionally supplemented with obstructions.