Field Trial of an ML-Assisted Phase-Sensitive OTDR for Traffic Monitoring in Smart Cities

We present a field trial of cost-effective distributed optical fiber sensing (DOFS) for traffic intensity estimation. The field trial uses a buried single-mode fiber installed along the university campus intended for communications and research experimentation. Acquired $\phi$-OTDR traces are concatenated and transformed into a space-time domain to form images. The resulting images are then processed by a MobileNetv2 model for car counting. Training is carried out by a set of collected images labeled by traffic information gathered automatically from camera videos processed with the novel YOLOv8 model. The results indicate that the implemented setup successfully counts heavy vehicles with high accuracy. The setup is also applied to data sets collected over entire days, showcasing its successful application for traffic monitoring in smart cities.