Deep Deconvolution for Traffic Analysis with Distributed Acoustic Sensing Data

Martijn Van_Den_Ende

Université Côte d’Azur, OCA, UMR Lagrange; Université Côte d’Azur, IRD, CNRS, Observatoire de la Côte d’Azur, Géoazur -
M. van den Ende, A. Ferrari, A. Sladen, C. Richard

Distributed Acoustic Sensing (DAS) is a relatively new technology which converts fibre-optic cables (like those used in telecommunication) into arrays of vibration sensors, recording continuous time series every few metres along the cable. Given how widespread fibre-optic cables are both on land and in the oceans, DAS hold enormous potential in numerous domains, including seismology, civil engineering, and marine sciences. One particular application that will be considered here, is that of vehicular traffic analysis.

Whenever a car drives past a DAS cable, it leaves a very distinct signal in the recordings, which can be used to identify, count, and track cars as they drive over the road. These analyses get more challenging when the cars are trailing closely in time (such as during rush hour), in which case the signals start to overlap. Using a self-supervised Deep Learning approach, we efficiently deconvolve the characteristic car signals from the DAS data, allowing for a greatly improved precision of subsequent analyses. By doing so, we are able to estimate the speed of the cars with a precision of 1-2 km/h.

Martijn van den Ende is an earthquake seismologist working at Géoazur / Université Côte d’Azur. His research interests span earthquake source mechanics, seismic cycle modelling, fibre-optic sensing, seismic array processing, and Deep Learning methods.