TensorFlow detection of traffic signal colors

[John Peters]

-Beep-tone on Green-

We are training TensorFlow to recognize the green traffic light and beep when the red light turns to green. It will run on a raspberry pI sitting on the dashboard. Of course it will only beep when the car is not moving and a red light turns green. It will work the same as it does in a Tesla. 

We have long traffic controls here in San Francisco and I tend to look around while waiting for the light to change to green. Sometimes I get beeped by the driver behind me. After training, we will run it on a Raspberry pi, smart camera and the coral USB accelerator.

[brooklin...@gmail.com]

1. Do we need to train on red and green images if we are only interested in detecting green?
2. Does it make since for it to ID a red light as green if we haven't trained on any red lights in the last training session?

3. So TensorFlow does not retain knowledge

[brooklin...@gmail.com]

I have learned the following:

++Training

Use non-biased training.

Split the data %80, %20

Don’t use the training data for testing

Run the training more than once.

The above is more of an art than a science.