Traffic Light Classifier

All questions answered

In the project notebook, all questions are answered. (There are two questions total.)

Standardize the input images

All input images (before they are classified) should be processed so that they are the same size.

One-hot encode all output labels

All labels should be a one-hot encoded vector of length 3. Ex. ‘yellow’ becomes: [0, 1, 0].

Create a brightness feature that uses HSV color space

Using HSV colorspace, extract a feature from a traffic light image that represents the level(s) of brightness in an image. This feature can help classify any traffic light image. A feature can be a list, array, or a single value.

Build a complete classifier

Using any created features, write a classification function that takes in a standardized RGB image and outputs whether a traffic light is red, yellow, or green as a one-hot encoded label.

Accuracy

The model must have greater than 90% accuracy on the given test set.

No red lights labeled as green

In the given test set, red traffic lights can never be mistakenly labeled as green.