click here to download project abstract
click here to download project base paper
ABSTRACT
Traffic sign detection and recognition has always been a concern as it has many applications in the near future. One of the most important being autonomous cars etc. this project is to use convolutional neural network algorithms to train the model with a data set of different traffic signs, test and validate with a high accuracy rating. This is a basic step which is to come. It can be used for highway maintenance. driver support exercises and more. The given data set has 43 different classes. The model’s accuracy for a given class is around 99(average) percent for signs that have a minimum of 500 images or more. The accuracy relatively drops to 80(avg) percent as the number of images drops to 100 or less. The reason for this is that the epochs(iterations) run are Ten. As the given data set is relatively huge the iterations take hours to train. The CNN algorithm is best for such images such as symbols as it trains a huge number of images before testing and putting in the grayscale feature makes it more efficient. This system can be introduced in people’s daily life as it can save people from giving warnings to people about the traffic signs. This can be introduced in every vehicle just like the parking assistance that has been introduced in today’s vehicles.