Enhancing safety and operational efficiency in railway systems ben-efits from robust AI-powered perception, particularly for reliable obstacle andpedestrian detection. However, the prevalent black-box nature of contemporarydeep learning models presents significant challenges for verification and trust,especially within safety-critical railway environments characterised by dynamicweather, illumination changes, and high speed, which create undesirable effectssuch as cluttered backgrounds, and motion blur, which may hinder the perfor-mance of computer vision approaches. This paper addresses the need for moretransparent models by proposing the application of Prototypical Part Networks(ProtoPNet) for interpretable obstacle and pedestrian classification within therailway domain. Experiments with the OSDaR23 dataset demonstrate that train-ing with a careful selection of data augmentation processes enhances key metricssuch as precision, recall and F1-score while yielding transparent results with vi-sually robust prototypes.