Traffic monitoring using distributed fiber optic sensing

1. A method for traffic monitoring using distributed fiber optic sensing (DFOS), the method comprising:

providing a DFOS system including:

a length optical fiber cable positioned along a roadway;

a DFOS interrogator system in optical communication with the optical fiber cable; and

an intelligent analyzer configured to analyze Dims sensing data received by the DFOS interrogator system;

operating the DFOS system produce Rayleigh backscattering light signal(s) from the optical fiber; and

processing the backscattered light signals to generate a waterfall plot in real-time; the METHOD CHARACTERIZED BY:

enhancing the waterfall plot image contrast;

extracting a vectorized representation of vehicle traces in the enhanced plot image;

generating real-time traffic information from the vectorized vehicle traces; and

outputting an indicia of the generated real-time traffic information,

wherein the waterfall plot image contrast is enhanced by one of a technique selected from the group consisting of: a convolution with structuring kernels, and a deep learning based contextual image prediction.

2. The method of claim 1 , wherein the convolution with structuring kernels comprises:

creating a bank of structuring kernels, covering a range of angles and widths; converting the image to gray scale;

convolving the waterfall plot with each structuring kernel and obtaining a convolution map for each;

computing a pixel intensity for each individual map; and

merging all convolution maps using the maximum intensity at each pixel.

3. The method of claim 1 , wherein the deep learning based contextual image prediction comprises:

Offline, one-time generating a training set comprised of a plurality of waterfall plots, each having different properties through the effect of a simulator; and train a neural-network model to predict any part of the waterfall plot given its surrounding context for a waterfall plot, applying the trained model to predict an enhanced output image.

4. The method of claim 2 , wherein the vehicle trace extraction includes:

performing image thresholding on the waterfall images, keeping selected trace snippets (seeds)

perform Hilditch thinning on the thresholded images to obtain skeletonized seeds;

perform a beam search to extend seeds along ridges of intensities above a predetermined threshold;

convert extended seed skeletons to vectorized polylines traces; and

reuniting trace snippets that belong to a same trace through the effect of colinear clustering.

5. The method of claim 4 , wherein the real time traffic information is generated by; processing the vehicle traces by linear approximation to deduce information for individual vehicles, wherein the individual vehicle information is selected from the group consisting of average speed over a time period, average speed over a roadway section, vehicle acceleration, and vehicle deceleration.

6. The method of claim 3 , wherein the vehicle trace extraction includes:

performing image thresholding on the waterfall images, keeping selected trace snippets (seeds);

perform Hilditch thinning on the thresholded images to obtain skeletonized seeds;

perform a beam search to extend seeds along ridges of intensities above a predetermined threshold;

convert extended seed skeletons to vectorized polylines traces; and

reuniting trace snippets that belong to a same trace through the effect of colinear clustering.

7. The method of claim 6 , wherein the real time traffic information is generated by: processing the vehicle traces by linear approximation to deduce information for individual vehicles a wherein the individual vehicle information is selected from the group consisting of average speed over a time period, average speed over a roadway section, vehicle acceleration, and vehicle deceleration.