Road-Speed Estimation Using Convolutional Neural Networks and Simulated ϕ-OTDR Traces

We investigate distributed fiber optic sensing and machine-learning-based image analysis for road-speed estimation. Synthetic phase-sensitive optical time-domain reflectometer (ϕ-OTDR) traces are generated by the simulation of random road features such as car density and speed. Consecutive ϕ-OTDR traces are stacked generating images that are submitted to a convolutional neural network (CNN) for classification. The evaluated CNN-based classifier exhibits high accuracy at sufficiently high car densities.