/traffic-cv

Toolset for determining physical state from traffic cameras.

Primary LanguageCSSMIT LicenseMIT

Traffic CV

Toolset for determining physical state from traffic cameras.

Build Status Coverage Status License GitHub contributors

Demo

demo.gif

Run Traffic Light Demo

node index.js

Then open up index.html in your browser.

Train Traffic Light Neural Net

node train.js

You must have training images in /images/training/green, /images/training/yellow, and /images/training/red.

Run the Trained Neural Net Through Test Suite

npm test

Will test the net against the training and testing data. Make sure to have categorized images in /images/testing.

Test Neural Net with a Specific Image

node run.js [path_to_image]

Example:

node run.js images/38-1531834988419.jpeg
[ 0.01618855582872857, -0.00072788746163907, 0.46764020897403424 ]
Traffic lights are 'red' with a 46.76% probability.

Capture Training/Testing Data

node index.js --capture

Will store captured images in /images/capture directory. You can then put those images in proper state directory in /images/training or /images/testing directory.