Onboard Vision Based System for Automatic Rock Detection on Rail Tracks

Railway transportation is a cornerstone of global infrastructure, ensuring efficient and reliable movement of goods and passengers. However, the safety and operational integrity of rail networks are constantly challenged by the presence of obstacles on the tracks, such as rocks. In this paper, we propose a novel approach to enhance rail safety by introducing an automated system that detects the presence of rocks on railway tracks. Leveraging an onboard camera mounted on the train, our method combines semantic segmentation and deep neural network classification to identify rocks in real-time. The experimentation, conducted on a novel dataset that we propose in this paper and that we make publicly available (composed of more than 6,000 images of railway tracks with and without rocks), confirms the effectiveness of the proposed approach.