FusionOptics: Optical-visual fusion framework for high-speed railway perimeter intrusion detection

Perimeter intrusion detection is essential for ensuring the reliability of high-speed railway automatic train operation (ATO) systems, where timely alerts are crucial due to the extended braking distances at high velocities. Although various sensing modalities have been explored, single-sensor approaches often struggle to provide robust performance under complex environmental conditions. Multi-sensor fusion has emerged as a promising alternative; however, it still faces challenges such as poor conflict resolution, static fusion weighting, and limited resilience to noise. To address these limitations, this paper presents an optical-visual fusion framework that integrates distributed fiber optic sensing with video surveillance. First, a spatiotemporal image-based vibration fiber optic signal recognition algorithm is proposed by integrating Sinc convolution, multi-scale feature fusion, and a channel attention mechanism. Second, a quality-gated decision-level fusion framework is proposed based on Dempster-Shafer evidence theory and image quality assessment. Finally, a long-range perimeter intrusion detection prototype system is developed and deployed in a real-world railway environment for practical engineering applications. Extensive experiments on a custom optical-visual fusion dataset demonstrate that the proposed optical-visual fusion framework effectively recognizes various intrusions and achieves a detection accuracy of 98.03%.