Detecting traffic lights and estimating the distance to them within given video using image processing and machine learning.
This project is parted into 4 phases:
Detection of source lights in an image using convolution with customized high-pass filter and local maximum.
Generating Convolutional Neural Network (CNN) to determine on the candidates from the previous phase whether a traffic light or not, using the the attention code output as labeled data for training, in order to optimize the training and reduce overfitting.
Estimation the distance of each traffic light using Struct From Motion and geometric calculations.
Integrating all the previous parts into a complete software product.