Modern railways feature increasingly complex embedded computing systems for surveillance that are moving towards fully wireless smart-sensors. Those systems are aimed at monitoring system status from a physical-security viewpoint, in order to detect intrusions and other environmental anomalies. However, the same systems used for physical-security surveillance are vulnerable to cyber-security threats, since they feature distributed hardware and software architectures often interconnected by 'open networks', like wireless channels and the internet. In this paper, we show how the integrated approach to security, privacy and dependability (SPD) in embedded systems provided by the SHIELD framework (developed within the EU funded pSHIELD and nSHIELD research projects) can be applied to railway surveillance systems in order to measure and improve their SPD level. SHIELD implements a layered architecture (node, network, middleware and overlay) and orchestrates SPD mechanisms based on ontology models, appropriate metrics and composability. The results of prototypical application to a real-world demonstrator show the effectiveness of SHIELD and justify its practical applicability in industrial settings.