Enhancing Safety by Obstacle Detection at Railway Level Crossings

Automatically identifying the presence of obstacles at railway level crossings is an important and not negligible feature in order to guarantee the safety of both people and goods, but also to guarantee a smooth mobility of road and rail traffic. Within this context, in this paper we propose a deep learning based video analytics algorithm, optimized to run on board of a smart camera, able to automatically detect the presence of people, vehicles, animals and generic obstacles on the railway at level crossing. The experimentation, conducted over a dataset of 50,000 images and on videos collected in the wild, confirms the effectiveness of the proposed system, which achieves a precision of 87.6% and a recall of 83.4%.