Contributers: Harpreet Virk, Dhruv Sabharwal, Yash Dixit
This project deals with building a crucial component of driverless cars viz. automatic traffic sign recognition using a Convolutional Neural Network Model (CNN). The model uses the Belgium TS dataset for this purpose. (http://btsd.ethz.ch/shareddata/)
After training the model for 50 epochs our best model accuracy turned out to be 91% on the validation data. This is indicative of the fact that from the 62 unique labels our model can accurately predicted the label class for 50+ labels. Majority of the incorrectly predicted labels still exist in the first half of the label set , i.e labels 2-15 . The reason for this is still their resemblance in terms of color, design and the same sign pattern.