Distributed Acoustic Sensor System for Intelligent Transportation using Deep Learning

We present DAS sample and 1D/2D CNN for vehicle type and occupancy classification

1. How to setup
   Step 1: Download and unzip preprocessed data in current directory: https://drive.google.com/file/d/1ore1g5sN8bUA7NvG5lvAGw9z6_LQlRuO/view?usp=drive_link
   

   or alternatively (not recommand)
   

   Download and unzip raw data (1) in current directory and run the following notebooks (2):

   • (1) raw data: https://drive.google.com/file/d/1RvyaRBf5PyBU4nVys5bn6OjHhCfGOT1c/view?usp=drive_link
   • (2) generate 058 (5p) txt data.ipynb and generate 058.5 (5p_to_1p) txt data.ipynb

   Step 2: Install pandas, numpy, and tensorflow
   Step 3: Keep in mind how the names of datasets in this repository are related to the paper:

   name in paper	pre-proccessed data in the repository	dataset description
   RC-60-Mix	058_5p_to_1p_X.txt 058_5p_to_1p_y.txt	Car 2 with 5, 4, 3, 2, 1 passengers
   AllCars-1p (= is now re-named to RC-60-1p)	026_X.txt 026_y.txt	The file contains signals from Car 1, 2, 3, 4, 5 with 1 passengers (a driver) but only Car 2 data is used for this individual testing dataset
   RC-60-5p	058_5p_X.txt 058_5p_y.txt	Car 2 with 5 passengers

2. How to use this repository
   

   We try to reproduce the testing results from Table III and Table IIV in our paper (named the same as this repository) so we named jupyter notebooks the same as the experiments in Table III and Table IIV.

   For example, in 5 way - 1d.ipynb, we trained a 1dcnn to classifiy exact number of passengers (5 way: 5 classes each of them has different number of passenger from 1 to 5).

3. Current reproducing results:
   

   	5-way	5-way	2-way	2-way	2-way +	2-way +
   Model of CNN	1D	2D	1D	2D	1D	2D
   Train:Test	67:33	80:20	80:20	80:20	80:20	80:20
   RC-60-Mix	0.81	0.93	0.896	0.98		0.96
   Ind. AllCars-1p (RC-60-1p)	0.297	0.68	0.46	0.62		0.52
   Ind. RC-60-5p	0.099	0.28	0.247	0.52		0.53